SiftKeyPre: A Vehicle Recognition Method Based on SIFT Key-Points Preference in Car-Face Image

Abstract

Vehicle recognition from images produced in roads bayonet provides important clues to solve vehicle crime cases. Its accuracy is not enough to meet the requirement in real conditions. We proposed a vehicle recognition method, SiftKeyPre, based on SIFT(Scale-invariant feature transform) key points preference for car-face images. Firstly, SiftKeyPre choices the SIFT key points following the DualMax algorithm to get a DualMax set. Meanwhile, Lowe set is defined as another one following Lowe algorithm. Secondly, we define a DL set under an intersection operation on DualMax set and Lowe set. For positive examples training images, we count the appearance times of each key point of DL set to compute the attention degree of each key point in base image. Finally, matching degree between the base image and a target image is evaluated with the attention degree of each matched points. SiftKeyPre method confirms a testing image based on its matching degree. Experiments results show that, under a given recall constraints, the precision of SiftKeyPre method is better than FLANN and Lowe. SiftKeyPre’s computational complexity is closed to that of Lowe. Comparing with other algorithms based on training, SiftKeyPre is of lower training intensity.