Lighttrack: system and method for online top-down human pose tracking

Abstract

A system and a method for pose tracking, particularly for top-down, online, multi-person pose tracking. The system includes a computing device having a processor and a storage device storing computer executable code. The computer executable code, when executed at the processor, is configured to: provide a plurality of sequential frames of a video, the sequential frames comprising at least one keyframe and a plurality of non-keyframes; for each of the non-keyframes: receive a previous inference bounding box of an object inferred from a previous frame; estimate keypoints from the non-keyframe in an area defined by the previous inference bounding box to obtain estimated keypoints; determine object state based on the estimated keypoints, wherein the object state comprise a 'tracked' state and a 'lost' state; and when the object state is “tracked,” infer an inference bounding box based on the estimated keypoints to process a frame next to the non-keyframe.