LiDAR-Assisted 3D Human Detection for Video Surveillance

Miquel Romero Blanch, Zenjie Li, Sergio Escalera, Kamal Nasrollahi; Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops, 2024, pp. 123-131

Abstract

This work explores 3D object detection using LiDAR technology, specifically focusing on pedestrian detection for video surveillance. While LiDAR is well-established in autonomous driving, its application in video surveillance is underexplored. We adapt state-of-the-art autonomous driving models for video surveillance, with CenterPoint being the top performer. Optimizing hyperparameters, such as voxel size and sweep merging, enhances pedestrian detection. Incorporating larger range data aids in generalization for video surveillance scenarios. This research demonstrates the feasibility of pedestrian detection in video surveillance and highlights open challenges related to domain adaptation and the high cost of high-resolution LiDAR sensors. Code: https://github.com/0Miquel/ OpenPCDet-video-surveillance.

Related Material

[pdf]
[bibtex]
@InProceedings{Blanch_2024_WACV, author = {Blanch, Miquel Romero and Li, Zenjie and Escalera, Sergio and Nasrollahi, Kamal}, title = {LiDAR-Assisted 3D Human Detection for Video Surveillance}, booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops}, month = {January}, year = {2024}, pages = {123-131} }