Abstract We developed a car counting system using car detection methods for both daytime and nighttime traffic scenes. The detection methods comprise two stages: car hypothesis generation and hypothesis verification. For daytime traffic scenes, we proposed a new car hypothesis generation by rapidly locating car windshield regions, which are used to estimate car positions in occlusion situations. For car hypothesis at nighttime, we proposed an approach using k-means clustering-based segmentation to find headlight candidates to facilitate the later pairing process. Counting decision is made from Kalman filter-based tracking, followed by rule-based verification. The results evaluated on real-world traffic videos show that our system can work well in different conditions of lighting and occlusion.
A Two-Stage Detection Approach for Car Counting in Day and Nighttime
2018-01-01
13 pages
Article/Chapter (Book)
Electronic Resource
English
A two-stage approach for front-view vehicle detection and counting at nighttime
DOAJ | 2016
|Nighttime Vehicle Detection for Heavy Trucks
British Library Conference Proceedings | 2015
|