Abstract This chapter proposes a multi-resolution hypothesis–validation structure to detect on-road vehicles in time. We remove the limitation of A. Broggio’s approach and build a simple and efficient hypothesis–validation structure which combines three steps. In the ROI determination step, we extract ROIs using lane markings instead of fixed regions. Following that, in multi-resolution hypothesis generation, the ROIs complement each other, and we utilize appropriate constraints to improve the search efficiency and reduce the computing burden. For validating vehicle hypotheses, the vehicle is represented with Gabor features, and then the feature vector of a vehicle is input for the SVM classifier. Based on boosted Gabor features, we propose a supervised learning approach to improve the performance. Experimental results show that our vehicle detector can improve both the detection rate and the robustness in real time.
Vehicle Detection and Tracking
2011-01-01
20 pages
Article/Chapter (Book)
Electronic Resource
English
Multiple vehicle detection and tracking
SPIE | 1997
|Vehicle and pedestrian detection and tracking
Tema Archive | 1990
|Vehicle detection device and vehicle anti-tracking method
European Patent Office | 2015
|Autonomous vehicle obstacle detection and tracking
Tema Archive | 1998
|