Research On Detection Of Two-wheeled Vehicle Based On Mixed Traffic Video

Traffic flow data analysis and research is an important part of the Intelligent Transportation System, which is self-evident for safe and convenient transportation. Vehicle detection research, as part of the Intelligent Transportation System, has also become a research focus and emphasis, and makes a lot of widely used achievements.The paper researches Two-wheeled Vehicle detection by mixing image processing and machine learning methods, combines research status of domestic and foreign and domestic traffic flow conditions. Proposed two methods to detect Two-wheeled Vehicle,which using template match and machine learning. Specific contents of this paper are:(1)This paper summarizes and researches detection technology of ordinary vehicles and two-wheeled vehicle at domestic and foreign. combined with the actual test scene. Analyze and compare the traditional vehicle detection techniques(such as magnetic coils, ultrasound and infrared sensors, etc.) with video-based vehicle detection technology(such as optical flow method, frame difference, background subtraction method) technology in limitations, convenience of installation and intuitive of data processing.(2)Applied a method that uses mean template of two-wheeled vehicle in foreground to detect two-wheeled vehicle. First, analysis common image denoising, such as median filtering, Gaussian filtering by experiments. The foreground with more complete information and less noise can be obtained by and operation between the several morphological processing foreground of frame difference and the foreground of Gaussian mixture model. Using edge detection method to obtain moving vehicle image edge information, and use the two-wheeled vehicles of foreground to construct the average template and feedback the template to the foreground to detect two-wheeled vehicle. Vehicle counting use centroid trajectory analysis for detecting vehicles.(3)Used another method that uses Mixed Gaussian Model and Ada Boost algorithm to detect two-wheeled vehicle. The first step is pre-detection by the difference of shape and size of common vehicles and two-wheeled vehicles. The second step is training positive and negative samples classifier using Ada Boost algorithm and feature descriptor(such as local binary pattern-LBP, histogram of gradient-HOG, Haar).At last, using classifier to detect two-wheeled vehicles in pre-detection windows. The last choose, using LBP feature to descript samples, depends on the training time of classifier and the accurate of detection. The experiments show that the proposed method can speed up the detection speed, and effectively reduce the false detection rate.