FORT - Fisheye Online Realtime Tracking
Paper -
Fisheye detection dataset - https://github.com/BenoitFaureIMT/CERIS_FishEye
Detection.py - Contains detection algorithm objects (YOLOv7)
Filter.py - Contains the Kalman filter logic
ReID.py - Contains ReID algorithm objects (ResNeXt50)
Target.py - Contains the definition of a track object
Utils.py - Utility functions
detect.py - Used to run the code -> Creates the detector -> Creates the tracker -> Reads video -> Feeds images into detector to get detections -> Feeds detections into tracker
Use the package manager pip to install the required modules.
pip install -r requirements.pypython detect.py --video <path to video> --show_results --save_results [other args]All arguments:
| Option | Description |
|---|---|
| --wait_screen | add a wait screen before tracking starts |
| --show_results | display results on screen |
| --save_results | save display results |
| --save_MOT | save targets to file in MOT format |
| --video | path to video file |
| --weights | path to YoloV7 weights |
| --filter_weight_path | path to weiths of NNFilter |
| --age_max | age max of tracklets |
| --alpha | alpha coeff for EMA |
| --IoU_threshold | IoU threshold for association |
| --cosine_threshold | Cosine threshold for association |
| --cost_threshold | Threshold between high and low score detections |
| --distance_threshold | Distance threshold for association |