TrackNet: A Triplet metric-based method for Multi-Target Multi-Camera Vehicle Tracking
David Serrano, Francesc Net, Juan Antonio Rodr\'iguez, Igor Ugarte
TL;DR
TrackNet is a modular multi-camera vehicle tracking method that combines detection, single-camera tracking, and triplet metric learning for improved IDF1 performance in traffic videos.
Contribution
The paper introduces a novel triplet metric learning strategy integrated with detection and tracking for multi-camera vehicle tracking.
Findings
Achieved IDF1 score of 0.4733 on AI City Challenge
Demonstrated effectiveness of triplet metric learning in track matching
Provided a modular framework adaptable to traffic video analysis
Abstract
We present TrackNet, a method for Multi-Target Multi-Camera (MTMC) vehicle tracking from traffic video sequences. Cross-camera vehicle tracking has proved to be a challenging task due to perspective, scale and speed variance, as well occlusions and noise conditions. Our method is based on a modular approach that first detects vehicles frame-by-frame using Faster R-CNN, then tracks detections through single camera using Kalman filter, and finally matches tracks by a triplet metric learning strategy. We conduct experiments on TrackNet within the AI City Challenge framework, and present competitive IDF1 results of 0.4733.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsVideo Surveillance and Tracking Methods · Autonomous Vehicle Technology and Safety · IoT and GPS-based Vehicle Safety Systems
MethodsSPEED: Separable Pyramidal Pooling EncodEr-Decoder for Real-Time Monocular Depth Estimation on Low-Resource Settings · RoIPool · Convolution · Softmax · Region Proposal Network · Faster R-CNN