| With the popularization of cars,the car ownership of the whole world is also increasing year by year,following from which the traffic pressure also challenges the existing traffic control system.Recently,artificial intelligence has developed rapidly.Many research institutes have begun to study more intelligent traffic control systems,among which the detection of vehicles in video is the most basic part.The traditional vehicle detection methods,which take advantage of such as inter-frame difference method and features for example Haar and HOG,can detect vehicle detection to a certain extent.But the test results in complex traffic environment are not very satisfactory.There are many reasons,including too much kinds of vehicles,occlusion,rotation and illumination changes.Using deep learning technology,we can solve some problems that traditional methods can not solve.And deep learning technology is also used in many research of vehicle detection.In this paper,we mainly study on vehicle detection based on deep learning.The main work is as follow:This paper proposes a method that using traditional method to detect coarse position of vehicles and using deep learning method to get exact.position of vehicles.This paper presents a system which has nearly real-time vehicle detection speed using a certain control mechanism to combine the vehicle detection method based on deep learning and the traditional vehicle tracking method.Collected surveillance videos from six different road intersection monitoring cameras in Tianjin and test these videos using the system proposed in this paper.And extract some processing results from them and compare them with the results of manual identification,which verifies the effectiveness of the proposed system in the complex environment of reality. |
