Vehicle Tracking in Low Frame Rate Scenes using Instance Segmentation
Allysa Kate M. Brillantes, Edwin Sybingco, Argel Alejandro Bandala, Robert Kerwin C. Billones, Alexis M. Fillone, Elmer P. Dadios · 2022
Multiple object tracking is a necessary component in traffic surveillance systems. It is a requirement for today’s embedded visual surveillance systems to process videos at low frame rates. This paper aims to track objects at different frame rates focusing on improving the tracking at low frame rates. In the proposed model, an instance segmentation model was used. The mask prediction, together with the IOU of previous tracks and new observations, was utilized for data association. The proposed method was evaluated using evaluation metrics such as HOTA, MOTA, and IDF1 and achieved better results in tracking objects at low frame rates.