I can characterize a project that my company has with one of our customers.
Imagine you are in manufacturing and your job is to provide inspections and spot repairs on a product at the end of the assembly line. Imagine that three to five different people do this at the inspection station and each inspector has a specific function.
- Check the Tire Pressure
- Inspect the Brake Lines
- Check Fluids
- Start the Engine
- Open and Close the Doors and Hoods
Stuff like that. Imagine that this takes about 30 minutes per vehicle. Now I know this is not the process for cars, that would be ridiculously slow. But what we have done with Amazon ML is that we’ve gotten video of a similar process and we are using various toolkits to identify people doing certain tasks with certain tools. So we can essentially monitor the inspection in realtime and make sure all of the actions are taking place. Then we can time the average time it takes.
At the end of gathering thousands of hours of video we can identify the most efficient or most quality effective types of inspections that take place on the variety of manufactured goods.
So another good analogy would be looking at the performance of pit crews in auto racing, or even special teams performance in football games. It’s all about training the ML to recognize the actions of the players in the video stream. We’re using Kinesis Video in combination with TensorFlow. Stay tuned for more news on this project.
