In a manufacturing plant, operators are required to remain within their designated work center area to ensure efficiency and adherence to operational protocols. However, during a 10-hour shift, some operators may spend excessive time outside their designated areas, exceeding an allowable limit of 8 hours. This deviation can lead to productivity losses, workflow disruptions, and potential safety concerns, necessitating a system to monitor and manage operator presence effectively.

Our initial intuition was to create a bounding box around a person using the YOLO model and track them with a tracker like DeepSORT. However, we noticed that when a person moved from Cam 1 to Cam 2, their ID became unstable and was not being carried over. At that moment, we realized that we needed a re-identification (ReID) logic to track individuals across different cameras. So, we tried multiple ReID models and implemented them to improve tracking. Here is a Repo with all the experiments we have done So far,

1. Experiments_before_sam2 : In this repo we have all the Experiments we tried out before Discovering Sam2
2. techdome-repo : In This we have 2 best code that were present in Experiments_before_sam2 folder
3. Map_builder : In this folder We have Code for website where user can draw map and mark office layouts, map camera with camera streams and can see the live view
4. Sam2 : Here we have Implemented using model named SAM2 (Segment Anything Model) Which is newly developed model by facebook capable of tracking person even in occluded enviornment with great efficiency.