CV

BoT-SORT

BoT-SORT

BoT-SORT (Aharon et al., 2022) extends ByteTrack with camera-motion compensation and a learned re-identification feature, raising MOT17 MOTA to 80.5 and HOTA scores significantly. The CMC module estimates background motion via sparse feature matching and corrects Kalman predictions; this matters on drones, PTZ cameras, and any unstable mount. FI Tech…

Definition

BoT-SORT (Aharon et al., 2022) extends ByteTrack with camera-motion compensation and a learned re-identification feature, raising MOT17 MOTA to 80.5 and HOTA scores significantly. The CMC module estimates background motion via sparse feature matching and corrects Kalman predictions; this matters on drones, PTZ cameras, and any unstable mount. FI Tech uses BoT-SORT for cross-camera person re-ID across mega-project CCTV grids — a worker entering Camera 4 and exiting Camera 7 ten seconds later gets stitched into one trajectory with the same global ID, enabling end-to-end dwell time and access-zone compliance analytics.