Research On Vision-based Combined Vehicle Detection And Tracking

In order to reduce traffic congestion,traffic accident rate and environmental pollution caused by energy consumption,combined with the existing intelligent vehicle(IV)technology,this paper proposes a new mode of travel—combined vehicle travel.Its working principle is uploading the navigation data of each intelligent vehicle to the intelligent transportation network system,which is responsible for allocating traffic resources,so that the intelligent vehicles with similar paths can be connected in sequence to form a fleet.The problem to be solved in this paper is how to combine intelligent vehicles with similar paths.In this paper,vehicle detection,license plate recognition and vehicle tracking algorithms based on vision aimed at the front intelligent vehicle are studied.The specific contents are as follows:(1)Vehicle detection for intelligent vehicles in the front.Firstly,the vehicle hypothesis area generated by shadows under the vehicle is adopted;secondly,the H component of the HSV color space in the vehicle hypothesis area is extracted and the color histogram is generated,and the color histogram is compared between the vehicle hypothesis area and the vehicle template to complete the preliminary verification of the vehicle hypothesis area;finally,the vehicle hypothesis area is verified again by HOG feature + SVM.Experiments show that the proposed algorithm can detect the intelligent vehicle in front of the vehicle better.(2)License plate recognition for detected intelligent vehicle.In order to achieve fast license plate location,a fast license plate location method based on color difference is proposed.In order to solve the problem of stroke interaction in some Chinese characters,a method based on contour feature and vertical projection is used to segment license plate characters.Aiming at the low recognition rate of traditional LBP features for characters,an improved LBP feature + template matching method is adopted to recognize license plate characters.A large number of experiments have proved the effectiveness of the license plate recognition algorithm in this paper.(3)Vehicle tracking for detected intelligent vehicle.The normal KCF tracking algorithm only extracts the HOG features of the target and ignores the color information of the target.In this paper,HOG features and CN features are used to train the relevant filters independently and calculate the response values of candidate image blocks,and the final position of the target is determined according to the weights of the two features.The scale of the target is estimated by references to DSST tracking algorithm.The experimental results show that the proposed tracking algorithm has good accuracy,real-time performance and robustness.