Answers are grounded in this paper and cite their sources.
Evaluates a pose-guided framework for reconstructing full-body poses from occluded views, suggesting implications for traffic safety.
This paper presents a pose-guided reconstruction framework that recovers full-body pose and silhouette from minimal lowerlimb cues visible through the gap beneath an occluding vehicle. The implemented and evaluated pipeline consists of Occluded Lower-Limb Pose Detection (OLPD), Full-Body Pose Inference (FBPI), and Pose-Guided Silhouette Reconstruction (PGSR), which detect visible lower-limb keypoints, infer full-body keypoints, and generate a silhouette probability mask. We also introduce Minimum Required Visibility (MRV) to quantify the minimum visible body ratio required to satisfy a predefined reconstruction criterion. Under the adopted IoU ≥ 0.80 and Dice/F1 ≥ 0.90 criterion, the lowest visible ratio among successful test samples was 12.87%, and the measured latency of the FBPI + PGSR reconstruction path was 8.676 ms per sample, excluding OpenPose BODY 25 inference, vehicle-mask prediction, keypoint filtering, and HSV. OLPD results should be interpreted as a mask-given synthetic upper-bound because the current implementation assumes access to the synthetic vehicle-layer mask produced by the compositing pipeline. The reported 8.676 ms should be interpreted as a local, hardware-unspecified measurement of the FBPI+PGSR reconstruction path only. It excludes OpenPose BODY 25 inference, vehicle-mask prediction, keypoint filtering, and HSV, and therefore does not establish end-to-end real-time performance. Human Silhouette Verification (HSV) is not implemented or evaluated in the current paper and is discussed only as a prospective future extension.
No takes yet. Share an insight, caveat, or question.
Yoo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: