ML Ops Cloud to Edge Solution for Industrial Machine Vision Applications

A survey of AI based machine vision technologies available in the market was conducted. It was found that there are many business models and solutions which can be categorized as: end-to-end solution, platform based offering and custom model solution. End-to-end solutions provide a complete hardware and software toolset but may have trouble integrating into existing infrastructure. A platform-based offering can be integrated into existing infrastructure but does not provide all hardware and software components required. Finally, custom solutions where the AI model is available as a plug-in for the use-case may also be procured.

The functional requirements for a camera-based system in process safety were identified and contrasted with existing solutions in the market. It was found that conventional systems were not able to fulfill these requirements mainly due to the lack of technical capabilities, such as performance in different lighting conditions, weather, backgrounds, and electrical classifications. With advancements in AI and its application to machine vision, technical capabilities of these systems are improving. However, further development is required to ensure these capabilities are reliable, resilient, and robust enough to be moved from a testing and academic environment to an industrial setting. In this presentation, we propose an edge-to-cloud AI based machine vision event-based detection and notification solution for industrial applications.