An architecture for distributing object detection tasks from video frames to workers in an edge network.
- Connect GPU ready workers to session
- Run admin to view object tracking output
- Have video file for object tracking task in the correct directory
- Run the server and begin transmission of video to workers
- Workers will return inferred bounding boxes to the frame
- Server will send the ordered frames to admin monitor to draw
- Server has tolerance for late/out of order frames
- Workers are chosen for a particular frame based on a preference system
- Any worker consistantly underdelivering or has issues with connectivity is punished
- Workers with very poor performance are banned.
Throughput of the network depends entirely on the collaberation of workers, reducing server strain and increasing its bandwidth for other complex tasks.