Front-view car detection using vocabulary voting and mean-shift search

摘要: Vehicle detection has received much attention over the recent years. In this paper we mainly address on of cars in front-view static images and focus local features that describe structural characteristics particular. Our method is based constructing visual vocabularies for car non-car objects (i.e., background). Extended speeded up robust are used using K-means algorithm. For a test image, e-SURF keypoints extracted possible background rejected order to retain by vocabulary voting strategy. The retained then scanned mean-shift algorithm find candidate bounding box car. extraction with searching technique repeated every scale down sampled image. set boxes found at mapped original image non-maximum suppression predict final system evaluated 25 distinctive vehicle classes 20 per class. Testing results show high rate 96.8% cars. flexible can be easily extended other views such as side or rear view modifying vocabularies.