Countor: Count Without Bells and Whistles

Andres Ospina, Felipe Torres; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020, pp. 600-601

Abstract


The effectiveness of an Intelligent transportation system (ITS) relies on the understanding of the vehicles behaviour. Different approaches are proposed to extract the attributes of the vehicles as Re-Identification (ReID) or multi-target single camera tracking (MTSC). The analysis of those attributes leads to the behavioural tasks as multi-target multi-camera tracking (MTMC) and Turn-counts (Count vehicles that go through a predefined path). In this work, we propose a novel approach to Turn-counts which uses a MTSC and a proposed path classifier. The proposed method is evaluated on CVPR AI City Challenge 2020. Our algorithm achieves second place in Turn-counts with a score of 0.9346.

Related Material


[pdf]
[bibtex]
@InProceedings{Ospina_2020_CVPR_Workshops,
author = {Ospina, Andres and Torres, Felipe},
title = {Countor: Count Without Bells and Whistles},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2020}
}