An Improved Feature Extraction And Matching Algorithm For Vehicles

With the development of society,the number of motor vehicles in China has increased.In order to safeguard the citizen's life and property safety,how to accurately locate and arrest illegal vehicles has become a problem now.Computer vision as an important branch in cutting-edge technology has been used maturely in vehicle license plate recognition.Because of high dimension and large memory requirement of the existed local feature description algorithms,there still not have effective solutions for the identification of clone care in expressway.To solve the above problems,this thesis has done the following work:1.An improved algorithm for vehicle feature points description was proposed in thesis after a lot of experiments.The vehicle face description set was constructed by extracting the improved ORB feature points and rBRIEF description algorithm from the front face of a variety of car models.Then the greedy search strategy was used to select the high variance and low correlation description dimension from the set,which were combined into the description model of the car face feature.This description model took up little space,and the accuracy of expressing the face feature was high.2.Aiming at the mismatching detection of feature points,the geometric constraint method based on prior knowledge was proposed Based on the distance between the matching points and the angle information,the matching constraint was constructed with the threshold set by the prior knowledge.In the case of predicting the rotation and scale transformation parameters between the images to be matched,the constraint method could effectively eliminate the false matching pairs.3.The algorithms proposed in thesis were applied to the highway bayonet clone car detection.Firstly,the preprocessing scheme based on the theory of dark channel prior principle was designed to enhance image quality.Haar-like feature and Adaboost classifier were used for vehicle target recognition and positioning.Secondly,due to the limitation of the memory of the highway access card,the vehicle face feature description model was used to express the feature point information after the feature points detected by the improved ORB algorithm,which made feature information occupy the pass card memory less than 1KB.Finally,in the phase of feature matching,the geometric constraint based on prior knowledge was used to eliminate the mismatch points,which improved the efficiency and accuracy of RANSAC algorithm.The experimental results showed that the algorithm of this thesis had good performance under both the same memory and the same characteristic points conditions.