Edge computing, deep learning and open-source Robot Operating System driving the evolution of machine vision

2018/04/17

San Jose

Manufacturers today are looking to transform traditional production systems as they struggle to addresses challenges associated with shortening product life cycles, rising labor costs and wide-ranging consumer demands. This has given rise to Industry 4.0, smart factory and Industrial Internet of Things (IIoT) developments, which are also propelling growing applications of machine vision technologies.

Alex Liang, product manager, IoT Solutions and Technology Group, ADLINK, thinks three critical trends will lead future developments of machine vision. With these trends come both new opportunities and challenges for machine vision.

First of all, with widespread applications of big data analytics, businesses demand higher data processing efficiency and are growingly making use of distributed computing, cloud computing, fog computing or edge computing. This is bringing changes to where and how machine vision is used. Machine vision systems have to make a breakthrough from the traditional concept of gauge, inspection, guide and identification (GIGI) to include the ability to perform computation and analysis and work in conjunction with other devices.

Furthermore, maturing IoT technologies have enabled a variety of devices to become connected. Communication across different devices via a network has become necessary. Therefore, communication solutions including OPC unified architecture (OPC-UA), Robot Operating System 2.0 (ROS 2), Data Distribution Service (DDS) and Message Queuing Telemetry Transport (MQTT) are increasingly important and will influence machine vision developments.

In response to shorter product life cycles, manufacturers endeavor to accelerate production to accommodate rapid market changes so they are keen to adopt easy and ready-to-use solutions to help them shorten the learning curve and speed up the introduction process. This is leading the development of machine vision software to gear toward simple, intuitive and user-friendly design.

Machine vision infused with innovations becomes instrumental to smart manufacturing

According to Liang, in the foreseeable future, machine vision will not only be used to perform quality inspection but it should also play an active role to allow robots to have human-like vision so that they can easily carry out loading, picking, gripping and packing operations with no need for an intricate guiding process. To achieve such purposes, machine vision systems have to incorporate the ability to collect, analyze and process large amounts of data in real time while supporting close communication with other devices. Therefore, how machine vision can keep up with advancing technologies such as edge computing, OPC-UA, ROS 2 and vision guided robotics (VGR) to fully support smart manufacturing requirements on high efficiency, high precision and low latency has become the next topic and challenge.

Take edge computing for example. The traditional machine vision design has separate camera modules and processing units (industrial PC). However, the rapid growth of data volume imposes increasingly challenging requirements on computational power so now the camera module has to turn into an edge-located computing node that can pre-process data and offload some of the burden from the processing unit.

OPC-UA, a machine-to-machine communication protocol for industrial automation, enables heterogeneous platforms or devices in a smart factory to communicate and exchange data. In the past, machine vision systems engage in communication with PLC, I/O or motion control equipment through various specific protocols or customized functions, making integration very difficult. The availability of OPC-UA will be able to resolve such problems.

The combination of ROS 2 and VGR equips robots or automated guided vehicles (AGV) with machine vision to enhance their efficiency and ability to work in synchronization. ROS is an open-source robotics operating system. The first generation ROS 1 is based on the TCP/IP protocol while the later generation ROS 2 is built on the UDP+DDS architecture and provides more powerful support for real-time data sharing between devices with robust security. Major robot manufacturers worldwide have all implemented support for ROS 2 using common SLAM, Navigation, Perception and Manipulation resources and algorithms. This not only enables problem-free communication across robotic systems but also builds a broad development platform for machine vision. Factories in the future will no longer have independent devices or work stations but instead can connect robotic arms, AGV and other machinery of different brands to realize smart manufacturing for wide-ranging production needs through the new VGR concept.

In addition to the above trends, the development of deep learning should also be considered when we try to figure out where machine vision technologies and applications are headed, commented Liang. Deep learning is not a new technology. It basically imitates the workings of the human brain to carry out recognition, decision-making and prediction through training using a neural network paradigm. Execution of deep learning tasks used to rely on high-performance CPUs but the costs ran high and the processing took a long time. Advancements in GPU technologies in recent years have allowed GPUs to process graphics with a much higher efficiency than CPUs. Deep learning leveraging the GPU's processing power has therefore become a technology with a high cost-performance value. The marriage between deep learning and machine vision can be expected to create far-reaching synergy in the future.

Looking ahead to 2018, ADLINK hopes to provide total solutions designed to enable quick introduction and optimal values for Industry 4.0, smart factory and IIoT applications. For this purpose, ADLINK will highlight edge computing, ROS 2 and deep learning as its focus R&D areas. The integration of these innovative technologies will complement the company's machine vision product portfolio, including smart cameras and image processing systems.