Research Papers

Automatic Traffic Data Collection under Varying Lighting and Temperature Conditions in Multimodal Environments: Thermal vs Visible Spectrum Video-based Systems

Filename 7C-Fu-FP.pdf
Filesize 849 KB
Version 1
Date added June 30, 2016
Downloaded 4 times/fois
Category 2016 CARSP XXVI Halifax
Tags Research and Evaluation, Session 7C, Student Paper Award Winner
Author Ting Fu, Joshua Stipancic, Sohail Zangenehpour, Luis F. Miranda-Moreno, Nicolas Saunier
Stream/Volet Research and Evaluation
Award/Prix Étudiant 3 Student

Slidedeck Presentation 

7C - Fu

Abstract

Vision-based monitoring systems using visible-spectrum (regular) video cameras can complement or substitute conventional sensors and provide rich positional and classification data. Recently, new camera technologies, including thermal video sensors, have become available and may improve the performance of digital video-based sensors. However, the performance of thermal cameras under various lighting and temperature conditions has rarely been evaluated at multimodal facilities including urban intersections, where road user classification is required. The purpose of this research is to integrate existing tracking and classification computer-vision methods for automated data collection and to evaluate the performance of thermal video sensors under varying lighting and temperature conditions. The evaluation is based on the detection, classification, and speed measurements of road users. For this purpose, thermal and regular video data was collected simultaneously under different conditions across multiple sites. Among the main findings, the results show that the regular-video sensor only narrowly outperformed the thermal sensor during daytime conditions. However, the performance of the thermal sensor is significantly better for low visibility and shadows conditions, in particular for pedestrian and cyclist data collection. Interestingly, the thermal video performs acceptably during daytime, with a miss rate around 5 %. This paper also shows the importance of retraining the algorithm on thermal data with an improvement in the global accuracy of 48 %. Moreover, speed measurements by the thermal camera were consistently more accurate than for the regular video at daytime and nighttime. The thermal videos are insensitive to lighting interference and pavement temperature, and solve the issues associated with visible light cameras for traffic data collection, especially for locations with pedestrians and cyclists.

Ting Fu, Joshua Stipancic, Sohail Zangenehpour, Luis F. Miranda-Moreno, Nicolas Saunier