Research On Moving Vehicle Detection Technology Based On Improved Differential Threshold Method And Deep Neural Network

With the continuous development of artificial intelligence and big data processing technology,the intelligent transportation system,which can implement extensiveness,multi-direction,accuracy and mass data processing,has been the shape of traffic system to come.As a hot topic in Computer Vision,the research of detecting vehicle target and distinguishing tracking technique has received frequent usage on intelligent transportation system.Not only can it solve the problems like investigating offending vehicles,tracking vehicles and collecting the vehicles' information,but also it can complete the task of managing traffic which necessarily invest a lot of manpower and time cost by programming intelligence algorithms.Nowadays,vehicle target detection is one of important techniques,especially for the unmanned.Unmanned vehicles can accurately avoid vehicles and ensure safe driving through this technique.Therefore,the research of detecting vehicle target and distinguishing tracking technique has important theoretical consequences and utility value.The main research contents and innovations of this dissertation are as follows:1.An improved differential threshold algorithm is proposed,which combines the background removal method for preprocessing based on the interframe difference threshold algorithm,and adopts a moving target clustering method to improve the detection accuracy.The dynamic threshold method is used to reduce the influence of light on the detection effect in the stage of image difference.The algorithm which has the traditional inter-frame difference threshold method has some advantages like its ease of implementation,high real-time detection and effectual anti-noise.And it can solve the "double-shadow" phenomenon of the inter-frame difference threshold method,and compared with the three-frame differential threshold method,the detection of the detection frame is better.From the experimental results of the detection,the proposed algorithm has two strengths of effectual detection and minor influence from light and noise.2.In this dissertation,a deep neural network structure is used to identify the target of detection.The new deep neural network structure Darknet-53 of YOLO3 is used for feature extraction,and the logistic function of YOLO3 is used to deal with the classification problem,it has the advantages of vehicle target recognition and fast discrimination.From the experimental analysis,it can well mark and track the moving vehicle target,and has achieved certain practical effects.The dissertation is divided into two major technical aspects: target detection and target classification.It can realize real-time detection and identification of mobile vehicles in the monitoring scene.Therefore,it can be applied to moving vehicle target detection and tracking.