| This paper studies vehicle detection based on monocular vision and discribes different methods for vehicle detection and vehicle tracking in the daytime and nighttime.It puts forward improved algorithms to solve actual problems acccording to the characteristics of different methods.In the daytime,It's important to extract traffic lane in vehicle detection.In order to solve the problem that the Hough transform can't extract bend lane and noise will greatly affect the accuracy of the result of curve fitting,this paper designs a symmetric search algorithm which use quartic polynomial to fit curve,this algorithm will increase the accuracy of the result of curve fitting.In order to reduce the processing time,this paper first extract the feature vector of the detection results by PCA,then calssify the feature vector by SVM.In the nighttime,the traditional methods will not only extract the tail lights but also the front lights and background lights which are noise in the vehicle detection,in order to solve this problem,this paper analyse and model the red halo and the dazzling light in the middle of the tail lights respectively,then segment out the red halo and the dazzling light in the middle of the tail lights from the frame using color threshold.The algorithm of Dempster-Shafer are used to verify the detection results of the tail lights.In the vehicle tracking,this paper designs a vehicle tracking algorithm based on motion model and tracking queue,the motion model estimates the position of the vehicle in the next frame and the algorithm searches the vehicle around this position by matching algorithm,the search result is used to verify the motion model and make up the tracking queue.Through the tracking queue the paper can track the vehicle smoothly.The result shows that when the algorithm works at the system of Windows 10 and 3.2 GHz CPU,the detection speed can achieve 20 frames per second and the algorithm can detect and track front vehicles accurately. |
PDF Full Text Request