Fast Timeline Based Multi Object Online Tracking

Abstract Fast state-of-the-art multi-object-tracking (MOT) schemes, such as reported in challenges MOT16 and Mot20, perform tracking on a single sensor, often couple detection, support only one kind of object representation or don’t take varying latencies update rates into account. We propose fast generic MOT system for use real world applications which is capable objects from different sensor / detector types with their respective rates. An SORT inspired online scheme extended by time awareness using timelines unifying concept. The supports object, filter modularizing generalizing the scheme, while ensuring high performance an efficient data-oriented C++-template-based implementation. Using proposed we achieve, comparable evaluation metrics, framerates up to ten times higher than fastest schemes publicly listed axis-aligned bounding-box MOT17 MOT20.