Ricoh s Second Stage Machine Vision: EMPOWERING DIGITAL WORKPLACES

Similar documents
Artificial Intelligence applied for electrical grid inspection using drones

Predicting the Future. The Downstream Benefits of a Predictive Maintenance Solution

Automation Technologies for Commercial Vehicle Safety Screening

Technical Paper. 1. Introduction. 2. Approach by Construction Equipment Manufacturer. Chikashi Shike Akinori Onodera Masamitsu Takahashi

Innovative Gauging. Best Practice Best Value. In-line Non-laser Non-contact. Robust. 2D/3D. Flexible. Reliable. Exact.

Hitachi Drone Platform Contributes to Urban Development

Solutions.

TSCM Cloud Services for Implementing the Global Mother Factory Center Concept

INNOVATIVE BRIDGE ASSESSMENT METHODS USING IMAGE PROCESSING AND INFRARED THERMOGRAPHY TECHNOLOGY

Drones: Without Intelligence, They re Just Flying Cameras. Aerialtronics, Aeryon and Drone America on Neurala Artificial Intelligence in Drones

EURAILSCOUT Inspection & Analysis

II. Presentation of the Project

Using People Analytics to Help Prevent Absences Due to Mental Health Issues

Revolutionizing Asset Management in the Electric Power Industry

3 STEPS TO MAKE YOUR SHARED SERVICE ORGANIZATION A DIGITAL POWERHOUSE

Research on Obstacle Avoidance System of UAV Based on Multi-sensor Fusion Technology Deng Ke, Hou Xiaosong, Wan Wenjie, Liu Shiyi

The Hybrid Automation Revolution

New Approach to Improving Site Operations of Distribution Centers with IoT Technology

AMERICAS Tel or Tel CHINA, SHENZHEN Tel

THE Industry Impact OF 5G. Insights from 10 sectors into the role of 5G

Optimizing the Manufacturing Line and Ensuring Flexibility to Double the Factory Productivity

Software Requirements Specification (SRS) Automated Pedestrian Collision Avoidance System (APCA)

Pavement Maintenance in Japan

Machine Learning Technologies for The Hospitality Industry

Internet of Things (IoT) Applications at Hong Kong International Airport

Revolutionizing Asset Management in the Water/Wastewater Industry

Digital-driven Financial Innovation and Hitachi s Involvement

FLEET SAFETY TECHNOLOGY. Risk Directory 2017/18

Teledyne DALSA Industrial Products. Embedded vision for industry

Logistics System Solution Expansion - From Sales of Items to Sales of Systems, From Automated Operation to Unmanned Operation -

AUTOMATED PICKING WITH AUGMENTED REALITY

Revolutionizing Asset Management in the Oil and Gas Industry

THE SMARTEST EYES IN THE SKY

APAS. Intelligent Systems for Human-Machine Collaboration

IoT Standard Pack to Facilitate Visualization and Remote Monitoring of Industrial Devices and Equipment

HOW TO USE AI IN BUSINESS

COMAU S FIRST AGV: A FLEXIBLE SOLUTION FOR THE SMART FACTORY

Developing the future of manufacturing and supply chain

New applications for Vision Systems in robot guidance and quality assurance

Establishment of spraying repair technology for concrete structures using drone

It s a NEW Day! A Framework for Digital Operations with an Analytics Foundation

PROVEN LEADERS IN 3D SCANNING AND INSPECTION FOR THE AUTOMOTIVE INDUSTRY

Accelerate Digital Transformation by Connecting Your Manufacturing and Supply Chain Data

Thinking Cities with HERE Technologies

Digital Workstyle Innovation to Improve Quality of Work

TCS Enables Connected Products Landscapes

Global Logistics Services for Value Chain Innovation

THE SMARTEST EYES IN THE SKY

INUDSTRY 4.0 SMART FACTORY

The Connected Industries Achievements, Challenges and Next Steps in Japan

THE SMARTEST EYES IN THE SKY

Product CerebroX.io All Rights Reserved

PROCESSING A SELECTION OF THE MOST REMARKABLE SOLUTIONS OF PRIMETALS TECHNOLOGIES FOR THE DIGITALIZATION OF

Comau Approach to Industry 4.0

Analytic Alphabet Soup: IoT, AI & ESP Big Data Analytics is a game changer in our Connected World

International Journal of Advanced Engineering and Management Research Vol. 2 Issue 1, 2017

S8911 Practical Application of Deep Learning in Smart Factory : Visual Inspection System of Semiconductor Laser

Cisco s Digital Transformation Supply Chain for the Digital Age

AVHS encompass a wide and complex variety of technologies, such as artificial intelligence,

Accenture Aerial Monitoring Solution. For Electric Utilities and Oil & Gas companies

Artificial intelligence A CORNERSTONE OF MONTRÉAL S ECONOMIC DEVELOPMENT

Enel is about to launch a global Call for startups to select innovative projects in the field of the Renewable Energies.

Technical Layout of Harbin Engineering University UAV for the International Aerial Robotics Competition

Additive manufacturing with NX

The future of agriculture Technologies shaping the industry

IBM Watson IoT Strategy

Panel: Choosing the right platform what is out there, which one to use

The 4.0 age of predictive simulation. WITNESS User Group, Italy, October 10th

DIGITAL DISRUPTION AT THE DOOR STEPS OF INDUSTRIAL ENTERPRISE. Digitalization 2.0 and Future of Enterprises

SERVICES OVERVIEW OF INFRASTRUCTURE MANAGEMENT SYSTEMS. Introduction

Intelligent Systems. For more information on partnering with the Kansas City Plant, contact:

SUPERCHARGING CAR SALES PLANNING WITH AI

Yale Lift Trucks Driven by Balyo

Background Information

The cargo facility of the future

Personally engaged in the entire transformation process, she attributed such progress to the revolution of digitization.

Capability White Paper Prescriptive Maintenance

SAP: Making digital transformation simple, agile, and powerful

Machine Monitoring and Indication Solutions

Nikon opens up about its strategic focus on Quality 4.0

CHAPTER 1 INTRODUCTION

Autonomous Battery Charging of Quadcopter

AEROTENNA TECHNOLOGY WHITE PAPER

Guide Superfast Broadband technology and the manufacturing sector

AUTOMATE YOUR ORGANIZATION

Disruptive Technologies Opportunities for the Asphalt Industry

From: Michael L. Sena To: Nick Bradley Ref: Safe Operation of Large Vehicles through Map-based ADAS Re: Proposed Article for Vision Zero

5G-based Driving Assistance for Autonomous Vehicles CMCC ZENGFENG

Realize the potential of a connected factory

AI in ITSM. Automate your IT to deliver great experience.

Executive Summary. Revision chart and history log "UNMANNED GROUND TACTICAL VEHICLE (UGTV)" Contract B-0068-GEM3-GC

SUSiEtec The Application Ready IoT Framework. Create your path to digitalization while predictively addressing your business needs

Next robot vision generation counts on embedded architecture and Touch & Automate

THE SMARTEST EYES IN THE SKY

Industry Innovations with SAP Leonardo: Mill Products Industry

HEXAGON METROLOGY WLS400A

OCTOBER Digital Supply Networks

Solutions for Dealing with Changes in Logistics Operations Environment

NEXT-GENERATION QUALITY SYSTEMS TO ENSURE FIRST-TIME QUALITY NDE 17 QUAL TESTNG

Νέες τάσεις για τη βελτιστοποίηση της εφοδιαστικής αλυσίδας. Business Development Manager, SCM Mantis Hellas

Transcription:

White Paper Ricoh s Second Stage Machine Vision: EMPOWERING DIGITAL WORKPLACES Ricoh has been providing automation and other labor-saving systems in many areas using machine vision. Our machine vision activities are now in their second stage and we are working on how we can elevate workplaces into a space for knowledge creation. Our system autonomously generates rules, understands the situation, presents optimal methods, and helps people at work to make decisions quickly. Ricoh aims to establish a smarter workplace through collaboration between intelligent machine vision and knowledge workers. 20/11/2017 Version 1.0.0

Table of Contents 1. Innovating Workplaces to Enable the Knowledge Creation... 2 2. Delivering Machine Vision to All Workplaces... 3 3. Ricoh s Second Stage Machine Vision Activities... 4 3.1. Second stage machine vision in social infrastructure... 5 Road Surface Inspection System Based on Stereo Cameras and AI... 5 Public infrastructure inspection systems using drones... 9 Autonomous flights of drones with a 3D vision system using an ultrawide-angle stereo camera... 12 3.2. Second-stage machine vision in Factory Automation (FA)... 15 Automatic random picking system using a stereo camera... 15 Image recognition and analysis technology for visual inspection based on machine learning... 17 3.3. Second stage machine vision in logistics... 19 Flow monitoring using stereo camera... 19 4. Tackling Second Stage Machine Vision with Open Innovation... 20 1 2017 Ricoh Company, Ltd.

1. Innovating Workplaces to Enable the Knowledge Creation With the rapid progress of digital innovation, workplaces are about to face a great change. Productivity has been improved, costs have been cut, and operations have been accelerated as we know it. The mode of corporate organizations has changed dramatically in just a few years, and so has the awareness of the people working there. Digital innovation has freed people from the constraints of time and space, bringing opportunities to create new value. The waves of innovation that permeate offices now reach the frontlines of many industries manufacturing, logistics, retail, healthcare, education, and more. A technology accelerating this movement is artificial intelligence (AI), which has gained huge attention recently. Particularly, deep learning based on the multi-layer neural networks, which is one of the technologies 1 of machine learning, has found applications in a variety of examples, enabling many people to recognize AI as something familiar. Expectations are rising for the dawn of the new era when machines (systems) are given human-like intelligence. As systems gain smartness, they will be able to do things that people have been doing or could not do in the past. A typical example is autonomous car driving. The technology not only reduces the burden on the driver but also increases safety, and is especially valuable in that regard. Efforts are underway to equip lighting and airconditioners with sensors, enabling automatic control of energy consumption according to the movement of people, thus creating a more comfortable environment. Workplace innovation, however, cannot be accomplished by merely introducing a 1 A scheme of finding characteristic patterns out of big data through iterative learning (statistical processing) 2 2017 Ricoh Company, Ltd.

system that features smartness. Innovation requires the knowledge creation, including the likes of value judgments and task setting, which cannot be replaced by AI no matter how excellent it is. Workplaces are reformed through the consistent knowledge creation by the people working there, and evolve into places for producing new value. The value of a smart system is determined by how much it contributes to the knowledge creation by people. Input Processing Output Collecting information Analyzing and processing Providing results to systems collected information and by optimal devices and people in optimal form converting it into intelligence Like sensory organs Like brain Like body response People perform a series of processes unconsciously in their daily tasks. A system must organically coordinate these processes to assist in the knowledge creation in a workplace. Knowledge creation in a workplace requires a wide scope of technologies that encompass input, processing, and output, as well as know-how in formulating an easy-to-use smart solution system. 2. Delivering Machine Vision to All Workplaces Ricoh has been nurturing technologies for optics, image processing, and electronic devices over many years. These technologies are integrated and can provide new value in the areas of machine vision. Ricoh has already established a prominent position in factory automation, (FA), automotive systems, security, distribution, and more. The concept is described in Ricoh s Machine Vision white paper. That white paper reveals Ricoh s commitment to establishing an intelligent technology that enables machines to not only operate according to instructions from people but also to quickly grasp the situation on behalf of a person and take appropriate action. Ricoh s machine vision facilitates the visualization of things that are not otherwise 3 2017 Ricoh Company, Ltd.

visible or automates a process that otherwise must depend upon human effort. With machine vision, Ricoh will allow people to take part in activities that will be increasingly productive and have higher added value. Our commitment will require a smart system that autonomously processes the entire scope of capture, analytics, and visualization. For Ricoh, this is the second stage of machine vision. Activities have already begun to enable a system to derive optimal judgment and action from vision-based information and to present it clearly. The system will repeat learning on its own, thus improving the accuracy of the information it presents. Primary element technologies of second stage machine vision Capture Analytics Visualize Stereo camera (including ultrawide-angle) Multi-spectral camera Extended depth-of-field camera Polarized camera Spherical camera Image processing Voice processing Natural language processing Data mining Electronic devices Printing Display Autonomous control 3D modeling Second stage machine vision is not just an alternative to the human eye. It will enhance people s value judgments and task setting, enabling advanced decisionmaking and reforming the workplace into a space for the knowledge creation. Activities in the second stage have already begun, covering a wide range of areas including social infrastructure, FA, logistics, and security. This white paper describes the status quo of Ricoh s second stage machine vision, and the value produced from these activities. 3. Ricoh s Second Stage Machine Vision Activities Ricoh believes that its activities in this second stage of machine vision will lead to smarter workplaces. The second stage has a vast range of applications, enabling 4 2017 Ricoh Company, Ltd.

solutions in a variety of fields. This document provides examples 2 in the areas of social infrastructure, FA and logistics. 3.1. Second stage machine vision in social infrastructure Road Surface Inspection System Based on Stereo Cameras and AI Roads enable traffic of cars and pedestrians, which is an important role as part of the infrastructure for social and economic activities and local lives. Yet road surfaces suffer constant deterioration and damage due to external forces such as heavy vehicle loads. If the roads are left unattended, the deterioration can result in unwanted incidents, such as a car crash due to the poor surface condition affecting the steering. To prevent such incidents, road surfaces must be inspected to find deterioration at an early stage and provide necessary repairs accordingly. Road surface conditions are evaluated according to three factors: the rate of cracks, the depth of ruts, and flatness (lengthwise bumpiness). Generally, the inspection requires a huge specialist vehicle provided with laser measuring instruments and digital cameras. The specialist vehicle, however, is very costly it is difficult to increase their number, and thus to increase the inspection frequency or expand the range. The vehicle cannot enter narrow roads, so inspection of community roads has been deficient. Three factors of road surface conditions: rate of cracks, depth of ruts, and flatness 2 Some are under development. 5 2017 Ricoh Company, Ltd.

Ricoh s Road Surface Inspection System features multiple stereo cameras; yet it is compact enough to be mounted on a standard motor vehicle. Unlike the conventional specialist vehicles, it can inspect community roads. The system allows automation of the work processes from capturing images by stereo cameras to creating inspection records, reducing the burden of inspection tasks. The greatest feature of the system is that the three factors of the road surface conditions can be inspected with only the stereo cameras 3. The rate of cracks, for instance, can be automatically determined. The cameras capture the images of the road surfaces and the system creates twodimensional images by combining feature points. Road surface monitoring system installed on a standard motor vehicle The automatic determination is based on machine learning 4, which is a type of AI. The created two-dimensional images are processed on a 50-cm grid, and the crack level (the number of cracks) is automatically determined by AI. A machine learning model has replaced human sight in the visual determination process, significantly improving efficiency and accuracy. 3 Click here to view a movie of the road surface inspection using stereo cameras. 4 Machine learning: Ricoh sets machine learning as one of the element development themes for enhancing image recognition technologies and has been studying it and developing its applications. 6 2017 Ricoh Company, Ltd.

Determining the rate of cracks by AI Where tires pass, ruts are caused by friction on the paved surface, sinking of the roadbed, and transformation of the paving material, for instance. To measure their depths, Ricoh s inspection system uses multiple stereo cameras to scan the road surface, automatically combines the scanned images along the width, and reveals the three-dimensional shape of the cross section of the road every 20 meters. Flatness of the paved surface influences the comfort of the drivers and passengers on traveling vehicles. Ricoh s inspection system continuously measures threedimensional shapes in the forward direction by capturing images by the stereo cameras and combining them. This eliminates the need for different measuring instruments or devices for measurement. We need only have the car-mounted stereo cameras scan the road surface and let the system measure the conditions. After measurement, the three factors of the road surface conditions need to be comprehensively evaluated to determine the need for repairs and the locations to be given priority. The evaluation should be based on the Maintenance Control Index (MCI). Ricoh s inspection system calculates an MCI quickly as it can measure the three factors on a single shooting drive. The MCI is indispensable for the comprehensive evaluation of the three factors. In the past, the management of community roads depended on visual measurement because the specialist vehicles could not enter and numerical data such as the MCI was not available. Ricoh s inspection system solves this problem. 7 2017 Ricoh Company, Ltd.

Ricoh s inspection system visualizes the road surface conditions by mapping the pavement conditions according to the measurement results. The mapped images are colored according to the MCI values, allowing the need for repairs to be quickly determined. Road surface condition mapping (example): Measured data is colored according to MCI values. A demonstrative experiment of the inspection system started in June 2016 with participation by the Ministry of Land, Infrastructure and Transport, Akita Prefecture, Semboku City, and Ricoh. A general-use vehicle was used to measure the road surface conditions for three kilometers of national, prefectural, and municipal roads in Semboku City, Akita Prefecture. The measurement was done twice, before the snow season (November 2016) and after the snow season (March 2017). The inspection system compared well with the specialist vehicle regarding MCI values for the three factors of road surface conditions. For the rate of cracks, high accuracy comparable to visual determination was attained through machine learning of many data samples. In addition, the demonstrative experiment confirmed how the paved surfaces are affected by snow, ice, snow clearing by snowplows, and snow-melting agents. The 8 2017 Ricoh Company, Ltd.

experiment revealed the need to grab changes regularly over time, for instance, and demonstrated the effectiveness of the measurement by a general-use vehicle. The results of the demonstrative experiment and the issues found were reported to the mayor of Semboku City on August 4, 2017, at the final debriefing meeting of the Consortium for the Demonstrative Experiment of the Road Surface Condition Inspection System*. Ricoh will continue to overcome technical issues and will strive to develop practical inspection systems that are accurate and easy to use. Ricoh is committed to implementing safe and secure road maintenance and management. *Consortium for the Demonstrative Experiment of the Road Surface Condition Inspection System: A joint project unit established by the Ministry of Land, Infrastructure and Transport, Akita Prefecture, Semboku City, and Ricoh for the demonstrative experiment of the road condition inspection system. Public infrastructure inspection systems using drones The aging social infrastructure has recently become a major problem for Japan s national and local governments. Many of the highways, bridges, tunnels, and buildings constructed in and after the high economic growth period (1960 s) have exceeded their service life of 50 years and need maintenance and repair. The Cabinet Office has been conducting a national project called the Strategic Innovation Creation Program (SIP). Under the theme of Infrastructure Maintenance, Renovation, and Management Technologies (supervised by NEDO: New Energy and Industrial Technology Development Organization), the Cabinet Office is promoting development of new technologies for the maintenance, renovation, and management of aged infrastructure. 9 2017 Ricoh Company, Ltd.

Supported by the program, Ricoh has been developing bridge inspection system with drones 5 that assist with close proximity visual inspection of bridges and systems to develop inspection reports. This has been done jointly with Tohoku University, Chiyoda Engineering Consultants, Japan AeroSpace Technology Foundation, and Tokyu Construction. It is an attempt to reduce the burden of inspection work on individuals; a camera is mounted on a drone to photograph items that require inspection. According to a 2014 survey by the Road Bureau of the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), Japan has about 700,000 bridges longer than two meters, 20% of which have been in service for more than 50 years. The percentage will continue to rise. In July 2014, MLIT started to impose obligations on local governments to inspect bridges every five years (MLIT Ordinance No. 39). In the implementation of the ordinance, however, both the dramatic increase in workload as well as the financial burden must be overcome. Inspecting a bridge with a spherical-shell drone Bridge maintenance and inspection processes are largely classified into field work and office work. First, the inspection staff conduct visual inspections on-site, chalk the damaged points, take pictures, and record findings in a field book. They then return to the office, summarize the data according to the records in the field book, and generate a detailed survey report on the damage, attaching photographs and sketches. The workload is burdensome, and report results can vary depending on the person. Bridge inspections are often more difficult due to both location and structure. When a bridge is difficult to access by an inspector, specialist inspection vehicles and 5 When a drone-mounted camera is used to inspect a bridge, the distance between the subject and the drone needs to be kept constant. The drone must enter a narrow space while avoiding beams and attached objects (such as cables, lights, and water pipes). So the drone is contained in a spherical frame about one meter in diameter; the frame comes in contact with the subject, keeping the drone at a distance and preventing direct collision. It was developed by Tadokoro Laboratory, Tohoku University. 10 2017 Ricoh Company, Ltd.

large-scale scaffolds are often required. Besides the inspection staff, traffic control staff may also be required depending on the situation. In these instances, it may also be necessary to address the influence of traffic control on the lives and economy of people in the surrounding neighborhood. Drones can be a solution to this issue. Drones with a photography equipment mounted on them can reduce the required number of people, cut costs, and ensure greater safety. Bridge inspection processes Additionally, Ricoh s bridge inspection solution greatly reduces the workload in office work. The latest image processing technologies help eliminate the troublesome and time-consuming tasks of capturing data from damaged locations and generating reports. From photographs (close-up images 6 ), a panoramic view of the entire bridge is automatically reconstructed in three dimensions (3D), and a 6 When inspecting a bridge, a spherical-shell drone performs a flyby across the inspection area to take close-up photos of all inspection points. 11 2017 Ricoh Company, Ltd.

development image is generated to provide a bird s-eye view of the points to be inspected. The system makes it easy to determine (detect) the locations and seriousness of any damage. In addition, Ricoh provides a document generation support tool for summarizing the results in a report, conforming to the inspection procedure recommended by MLIT. Automating the report generation process will not only reduce the workload but also eliminate the variation in report quality, reducing the number of man-hours required dramatically. Extracting damage points from photographed images and creating a survey report Beyond bridges, drones within a spherical shell can be safely used where maintenance and inspection by individuals poses a risk e.g. public structures like tunnels, super skyscrapers, and plant facilities. Ricoh will continue to develop the automation of the report generation process, including the efficient identification of damage that occurs in many different forms. Our goal is to provide a total solution for infrastructure inspection. Autonomous flights of drones with a 3D vision system using an ultrawideangle stereo camera Drones are expanding their applications in business areas thanks to rapid improvements e.g. increased structural reliability, higher environmental resistance, and enhanced navigation precision. As more and more drones are put into practical use, people expect them to fly beyond the boundaries of conventional physical restrictions. Demand has arisen for drones to undertake a stable flight outside the 12 2017 Ricoh Company, Ltd.

scope of the operator s view or without access to global positioning systems (GPS). Currently, most business-use drones require operation by an expert operator. However with the expansion of applications, operator shortage is anticipated. This is another reason why autonomous flight capability is required. GPS does not help in complex topography because errors in positioning can be in the order of meters. When drones are used inside a building or tunnel, GPS signals are unstable or unavailable in the first place. Even in an open space, GPS navigation is unsuitable for flight due to electromagnetic interference. Through joint research with the University of Tokyo and Blue Innovation, Ricoh has solved these issues and enabled autonomous flights and the avoidance of obstacles by drones based on a 3D vision system. Combined with an inertial measurement unit (IMU 7 ), our proprietary 3D vision system, equipped with an ultrawide-angle stereo camera, allows a drone to avoid obstacles autonomously and fly stably without an operator. Through coordination between a wide-view eye that captures the surrounding conditions and the semicircular canals that produce the sense of balance during flight, the drone autonomously flies in a non-gps environment, accurately tracing the path predetermined by the user. For the autonomous flight system, the University of Tokyo has developed flight controls, and Ricoh has developed the 3D vision system 8. The drone knows where it is using the ultrawide-angle stereo camera of the 3D vision system, measuring its distance to the surrounding objects in a wide view and forecasting its movement in reference to characteristic points and their relative distances. This technology has enabled stable autonomous flights where GPS signals are unavailable. The wide viewing angle of the camera allows the drone to keep the characteristic points in sight even when its attitude changes, thus forecasting its own movement stably. 7 A sensor that detects angular velocity (gyro) and acceleration 8 RICOH 3D Vision System Drone (Movie here) 13 2017 Ricoh Company, Ltd.

Although the path is predetermined by the user, unexpected obstacles may exist on the path. The 3D vision system generates a 3D map around the flight path concurrently with movement forecast, so the drone can detect and automatically avoid unexpected obstacles. The 3D vision system is unique in that it handles a series of processes in real time 9 inside the camera unit, from photography through to positioning forecasting and generating 3D maps. This is a new technology produced by the fusion of excellent optical design technology, advanced image processing technology, and elaborate systems control. Autonomous obstacle avoidance based on 3D vision system Ricoh s 3D vision system allows drones to be used without an operator both indoors and outdoors. It autonomously avoids obstacles whilst flying. Ricoh will continue to enhance the smartness of the 3D vision system, expanding the possibilities of drone applications. 9 Fast processing is attained by implementing multiple processes on an FPGA (Field- Programmable Gate Array) package, an integrated circuit configurable to specific design objectives. The processes include image processing, such as characteristic point extraction, and parallax operation for the stereo camera. 14 2017 Ricoh Company, Ltd.

3.2. Second-stage machine vision in Factory Automation (FA) Automatic random picking system using a stereo camera Ricoh has developed a fast and precise stereo camera primarily for industrial applications. A stereo camera can capture the depth of a subject. Developing it requires advanced calibration technology, parallax operation technology 10, precision parts mounting technology, and more. Quality requirements are acute in industrial fields, and developing a module for them requires particularly high skills and ample experience. Since the latter half of the 1970 s, Ricoh has been enhancing automation of its own production facilities using FA and accumulating technical know-how in this area. We have shared some of the results outside the Company e.g. machine vision, and have collaborated with other FA manufacturers. Ricoh has been contributing to the improvement of office productivity (Office Automation - OA) with multi-function printers (MFPs) and single-function printers and, at the same time, has attained many achievements in FA. FA is prevalent in manufacturing, but some processes still depend on human labor. Random picking is one of them. Parts that are randomly placed on a rack need to be individually transferred to a part feeder or an assembly line. Their conditions must be quickly determined while they are on the rack so that they can be picked up efficiently and properly arranged. Conventional automatic picking systems used multiple devices and robots, resulting in poor space efficiency and high cost. Many of their cameras were 2D they had a low recognition rate and were not reliable enough to replace human operators. 10 Calibration is the adjustment for detecting and measuring subjects to be photographed, and parallax operation is the processing to recognize the positions and shapes of subjects to be photographed. 15 2017 Ricoh Company, Ltd.

Then came a projector-based 3D vision system, but it was too large and slow 11 to readily fit on production lines. The picking systems based on the RICOH SV-M-S1 industrial-use stereo camera, is the solution to this difficult issue. With the stereo camera capturing the subject in real time in 3D, the picking systems can quickly recognize the shapes and orientation of parts that conventional cameras cannot. Collaborating with a robot manufacturer that has worked with us before, Ricoh has developed the picking robot into a system that maximizes the use of the fast and precise stereo camera. Before implementation, Ricoh Industrial Solutions, an experienced FA systems integrator within the Ricoh Group, can now propose an effective system from the viewpoint of the optimization of the whole process in collaboration with the customer. The system has already been introduced to the production lines of multiple customers, contributing to improved production efficiency. At this dawn of the IoT age, the functions required for the picking system are about to change. The movement for PLM (Product Lifecycle Management), a concept for centrally managing the entire lifecycle of a product, has become active internationally. The conventional system based on batch Robot picking system processing on the part feeder is incapable of answering the needs sufficiently. Ricoh s random picking system uses a stereo camera to recognize individual parts, and respond to such requirements flexibly. By developing smart system technologies that solve the issues of FA and providing customer-oriented solutions, Ricoh will continue to contribute to production innovation. 11 The projector-based system uses the phase shift method for 3D measurement, which requires extra processes: pattern projection on the projector and dot data processing on the camera. 16 2017 Ricoh Company, Ltd.

Image recognition and analysis technology for visual inspection based on machine learning The parts inspection process is considered the last hurdle in implementing FA in production processes. Thanks to the recent improvements in photography technology and image recognition technology, the applications of machine-visionbased visual inspection systems have expanded. Nevertheless, their capabilities have not been fully exploited. The delay is the difficulty in telling the difference between good and defective products. In addition to damage and flaws, many factors are included in making a comprehensive judgment based on a variety of conditions. When the boundary between the good and defective products is ambiguous, the judgment must be based on knowledge and experience. Still, results tend to vary when inspections are based on human labor. Depending on the parts, inspection tasks can burden the inspection staff (excessive use of eyesight and transfer of heavy objects). Automation of the inspection processes is a must in improving production efficiency, reducing costs, and effectively using staff (reassigning people to higher added value tasks). To provide a solution in this area, Ricoh has been conducting research and development into Image Recognition and Analysis Technology for Visual Inspection based on machine learning. Ricoh has been studying machine learning as an element technology for image recognition. One result is anomaly detection based on a semi-supervised learning method. This method lets the machine learn correct values (labels) only, so that anything that contains any other data is detected as an anomaly. Accuracy increases with the iteration of judgment (iterative learning), and the data quantity (the number of sample images) can be smaller than the generally used supervised learning (based on data sets in which all data items are labeled). Semi-Supervised learning can be enhanced by applying image processing 17 2017 Ricoh Company, Ltd.

algorithms 12 that calculate characteristic quantities from the photographed data of the parts. Using the enhanced method, Ricoh is developing a precision visual inspection system that automatically detects defective products. Our ample knowhow in image recognition contributes to the improved precision of machine learning. In general, defective parts are far fewer than good parts, so it is unrealistic to prepare learning samples of all defect types. Thus, the semi-supervised learning method is suitable because it is based only on the characteristic data of good products. This method can detect unknown defects, and has high precision in inspecting parts where the shapes of good products vary greatly (parts in the preliminary processing stage, for instance). The image recognition and analysis technology for visual inspection based on semisupervised learning received an award for excellence over two consecutive years (2014 and 2015) at the Visual Inspection Algorithm Contest, sponsored by the Technical Committee on Industrial Application of Image Processing and the Japan Society for Precision Engineering. Machine learning can be applied to many fields. Its value is increasing at this dawn of the Big Data age. Ricoh will continue to improve the learning precision and Function blocks of visual inspection system Sample detection result (Visual Inspection Algorithm Contest 2014) 12 SIFT (Scale-Invariant Feature Transform) to extract characteristic points from images (boundaries and edges) and SURF (Speeded-Up Robust Features) to calculate characteristic quantities (information robust against rotation, scale, and illumination) 18 2017 Ricoh Company, Ltd.

processing speed, aiming to take image recognition and analysis technology to an advanced level that compares to human judgment. 3.3. Second stage machine vision in logistics Flow monitoring using stereo camera The advancement in machine vision has enabled mass image information to be used in a variety of fields, but how to obtain useful information is another problem. It can be solved by the visualization technology, which analyzes the mass information as necessary and presents the results as visual information. Particularly, the technology to visualize dynamic information in real time assists quick decisionmaking, helping people discover things that are difficult to recognize in the visual inspection of objects or the investigation of static images. In a warehouse, for instance, there is a need to observe the motion (flow) of people after changing the shelf layout to see if the change was appropriate. Ricoh has developed a flow monitoring system using stereo cameras, and uses it to optimize the layout in the logistics warehouse of Ricoh Europe SCM B.V. (Bergen op Zoom, The Netherlands). Visualizing the flow in a warehouse (heat map) 19 2017 Ricoh Company, Ltd.

The flow monitoring system has multiple stereo cameras to monitor the conditions of the transport work in the warehouse in 3D in real time, and displays the results visually. The information on the workflow where people or loads tend to concentrate, for instance is classified and shown in shades of different colors, allowing people to grasp the situation at a glance. The obtained information can be analyzed from a range of viewpoints to figure out a layout that maximizes efficiency and minimizes workload. In the past, flow monitoring was done as fixed-point observation by a video camera. The fixed-point method required a long time to review, and things tended to be overlooked when the recording was fast-forwarded. The stereo cameras featured in Ricoh s flow monitoring system precisely capture and visualize the information, such as positions and flows, in 3D, enabling quick and appropriate decision-making. The visualized images can be analyzed from multiple angles under different conditions as necessary to determine the optimal layout. The application of the flow monitoring system is not limited to warehouses. In an office, for instance, it can be used in an ecology system for controlling energy according to the flow of people. It can also be used in a security system that issues an alarm upon detection of an anomaly in the incoming and outgoing flows of people. Ricoh will continue to further expand the applications, developing the monitoring system into a smart decision-making support system with excellent capabilities, including automatic analysis of the results obtained from monitoring. 4. Tackling Second Stage Machine Vision with Open Innovation This document has introduced Ricoh s products in the second stage of machine vision, including those under development. With each one, the developers look directly at how to increase the intelligent productivity of people, rather than merely seeking automation or labor-saving. Ricoh s attempts to make smart systems has 20 2017 Ricoh Company, Ltd.

just begun. Performance is being improved steadily through on-site verification and other efforts. Ricoh has accumulated numerous technologies over many years, not to mention the optical, image processing, and precision mounting technologies, and they are about to bring a shift in the status quo as a destructive innovation 13. Yet Ricoh will not conduct these efforts on its own. As seen in the examples introduced in this document, many of our development tasks are conducted through collaboration with research facilities, with multiple companies, and even with our customers. We are striving to create new value by bringing about innovation through the fusion between Ricoh s proprietary technologies and know-how and the technologies, knowledge, and know-how of our partners. Keep your eyes open for Ricoh s second stage machine vision, which is based on open innovation to establish smarter workplaces. 13 Using innovative technology to bring dramatic changes to the existing business models and industrial structures 21 2017 Ricoh Company, Ltd.

Revision history Ver 1.0.0 2017/11/20 First edition 22 2017 Ricoh Company, Ltd.