VPULab participation at AI City Challenge 2019

Elena Luna, Paula Moral, Juan C. SanMiguel, Alvaro Garcia-Martin, Jose M. Martinez; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 343-352


In this paper, we present an approach for Multi-target and Multi-Camera Vehicle Tracking and another approach for Vehicle Re-Identification (ReID) across multiple cameras. We evaluate both approaches over "CityFlow: A City- Scale Benchmark" participating in Track 1 and Track 2 of 2019 AI City Challenge Workshop. The proposed tracking approach is based on applying detection and tracking of multiple moving vehicles for each camera. Afterwards, we cluster such results (detections of vehicles) obtained from multiple cameras with overlapped fields of view. The clustering is based on appearance and spatial distances on a common plane for all camera views. The optimal number of clusters is obtained by using validation indexes. Then, a spatio-temporal linkage of the obtained clusters is performed to obtain the trajectories of each moving vehicle in the scene. We tested different combinations for the input of the proposed approach (detector and tracker) and provide sample results for selected scenarios of "CityFlow: A City- Scale Benchmark". The proposed re-identification system is based on the combination of adapted deep learning feature embedding representations and a distance metric learning process. We also include the vehicle tracking information provided by the "CityFlow: A City-Scale Benchmark" in order to improve the results. We tested different combinations of features, metric learning and the use of tracking information and provide sample results for the CityFlow- ReID dataset.

Related Material

author = {Luna, Elena and Moral, Paula and SanMiguel, Juan C. and Garcia-Martin, Alvaro and Martinez, Jose M.},
title = {VPULab participation at AI City Challenge 2019},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}