The Democratization of Machine Vision in the Industrial Sector

Subscribe To Download This Insight

2Q 2018 | IN-5139

Microsoft announced Project Brainwave and Project Kinect for Azure at the Microsoft Build 2018 conference. Both solutions are targeted at bringing machine vision to the industrial sector. This ABI Executive Foresight explores the main approaches adopted by different players and dives into the drivers and momentum behind machine vision in the industrial sector.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


Machine Vision Is No Longer Expensive to Deploy 


Due to the increasing levels of automation across different verticals, machine vision has become a critical component in machinery and operation workflows. Traditionally, machine vision is implemented based on computational-intensive approaches, such as edge extraction, line labeling, and polyhedral modeling. While machine vision has been used in quality control and sorting and assembly lines, the implementation is generally expensive and requires heavy upfront investments and a lengthy commissioning time.

In recent years, the emergence of cost-efficient hardware and quality accessible open-source software frameworks and tools have since lowered the barrier to entry significantly. Cloud-based deep learning algorithms, especially Convolutional Neural Networks (CNNs), are self-taught algorithms capable of finding and extracting unique patterns based on large amounts of data. Cost-efficient hardware, namely chipsets, cameras, and sensors, enable Artificial Intelligence (AI) inference to be performed at the edge, reducing the need for constant connectivity and bringing ultra-low latency into machine vision processes.

Microsoft and Intel Joining Forces in Machine Vision


One of the areas that will benefit greatly from the development of machine vision is industrial robotics. Commercial robotics have been reaping the benefits of vision-based Simultaneous Localization and Mapping (SLAM) technology, as chronicled in a previous ABI Executive Foresight, “Robotics and AI Evolution in the Chinese E-Commerce Fulfillment Market.” However, industrial robotics have been slow to adopt machine vision due to the lag in technology adoption in the industrial sector, and this has not gone unnoticed.

During the Microsoft Build 2018 conference, Microsoft announced two major machine vision initiatives. First, Microsoft launched Project Brainwave with Intel. Leveraging Intel’s Field Programmable Gate Array (FPGA) solutions, developers can train their image dataset using Microsoft Azure’s deep neural networks in the cloud, before implementing the solution at the edge. According to Microsoft, FPGAs outperforms Graphic Processing Units (GPUs) when there are custom data types or irregular parallelism. This enables FPGAs to support a wide range of machine learning algorithms for real-time live data streams, such as machine translation, computer vision, and video analytics.

Secondly, Microsoft is bringing back its popular Kinect camera. While Kinect made its name in the gaming market with Xbox, the camera module gained significant popularity within the robotics developer community due to its price-performance ratio. When Microsoft pulled the plug and stopped producing Kinect, it was believed that Microsoft had decided to exit the market, which has been taken over by Intel’s RealSense and NVIDIA’s Jetson platform.

Interestingly, Microsoft took a different path by launching a camera module with integrated cloud-based AI capabilities. Instead of having localized processing, Microsoft has migrated all data processing to its Azure cloud and houses the camera module with a Time of Flight (ToF) sensor for depth sensing, a Red, Green, Blue (RGB) camera, a 360° microphone array, and an accelerometer. This reduces both the size and the cost of the module. Microsoft believes the solution will be applicable to security (home surveillance camera), healthcare (automatic material handling), and manufacturing (autonomous mobile robot), where the hardware platform is constantly connected to the power supply and to the cloud.

Machine Vision to Gain Prominence in Industrial Robots


Meanwhile, many startups have been offering machine vision solutions targeting the industrial sector. Berlin-based micropsi industries is working with ABB, Universal Robots, and Franka to bring its machine vision software to industrial robots. Different from the cloud-based approach taken by Microsoft, micropsi industries opts for on-premise servers. Manufacturers are generally reluctant to host their business-critical data on a public cloud. So, micropsi industries’ solution is designed to perform training and inference of CNNs and Multilayer Perceptron (MLP) feed-forward networks in on-premise servers, using data collection from industrial cameras and force or torque sensors.

Another startup, Veo Robotics, uses machine vision to address the safety aspect of industrial robots, by making a safer robot using Three-Dimensional (3D) sensors and AI. In terms of machine vision software,, a startup by Andrew Ng, specifically names manufacturing as its target market. aims to help manufacturers develop a powerful inspection and quality assurance solution based on data collected from different sensors in the factory.

Solutions offered by Microsoft, micropsi industries, and Veo Robotics are by no means game changers, but it signifies the democratization of machine vision. The increase in machine vision deployments will certainly shore up the market for camera shipments in industrial robotics. ABI Research estimates that camera shipments specific to industrial robots will increase from 79,000 in 2018 to more than 400,000 in 2026. This takes into account shipments of embedded, standalone, and multi-camera systems with controllers, for both Two-Dimensional (2D) and 3D camera. Given the growing acceptance of collaborative robotics arms and autonomous mobile robots, machine vision will continue to play a huge role in making the industrial processes more efficient, collaborative, and intelligent.