Object Tracking using Optical Recognition
My friend, a patent lawyer, asked me to review a patent his firm was working on. After providing feedback and (ahem) identifying opportunities for improvements we thought it'd be interesting to build a prototype.
The patent concerns itself with tracking various pieces moving through a factory where it's important that
- Pieces disassembled from a larger piece retain the lineage from the master piece
- No markup, sticker, printing of any disassembled piece can take place
- During disassembly, pieces can be rotated in any way
- Various inspections of the smaller and large pieces takes place while moving around on the conveyor belt
- and a host of other requirements
Using a webcam and lego we built a prototype simulating the environment and disassembly environment. Software used various vision and object recognition libraries, as well as a custom inference and backend.
The prototype worked so well that the project has been funded and will be developed and deployed in the factory of a beta customer.