GeoComp demos

People re-identification in Lidar-based tracking

This video demonstrates moving pedestrian detection and trajectory extraction in the 3D space based on foreground regions detected in Lidar point clouds. We have proposed an on-line model for short and long term person tracking, and extracted relevant features for person re-identification from the Lidar data under challenging outdoor conditions with low point density and multiple occlusions (back to the surveillance demo page).

Demonstration video of multi target tracking. Solid coloring marks identified objects, while stripped coloring refers to objects avaiting identification. Identified object should receive the same color during the whole tracking period. Identification is online, and does not utilizes the camera view. Color map of the recognized people in the top right corner has been prepared in post processing and only serves easier verification for the demo observer.

Reference:

Cs. Benedek, ”3D People Surveillance on Range Data Sequences of a Rotating Lidar,” Pattern Recognition Letters, Special Issue on Depth Image Analysis, 2014