Research On The Application Of Candidate Frame Algorithm In Image Vehicle Detection

The proposals algorithm is an effective method to improve the efficiency of object detection and calculation in the near future.The proposals algorithm can select a candidate window that may contain a object in the window.Compared with the sliding window,the number of object search windows is reduced,and the accuracy of the window is improved.The proposals algorithm is also called a proposals recommendation algorithm,a object algorithm,a proposals sampling algorithm algorithm,and the like.The proposals algorithm instead of the sliding window method is applied as a preprocessing algorithm in the object detection.The main work of this article is as follows:(1)Manually annotate the vehicle data set according to the Qingdao traffic surveillance video.The image size is 1000.Vehicle data sets were acquired under different weather conditions,lighting conditions,and road conditions,including 23 shooting junctions at different time periods.(2)Selecting the proposals algorithm's representative Edge boxes algorithm and Selective search algorithm,as well as the emergence of the earliest Objectness algorithm to study,by calculating the recall rate and calculation efficiency of the three algorithms in the vehicle data set,the three algorithms Evaluation analysis,according to the scene and lighting conditions,selected shadow scenes,congestion scenes,night scenes to further compare the performance of different algorithms.Under the consistent evaluation criteria,the three proposals recall rates are: 0.86,0.36,and 0.53,respectively,and the performance is different.(3)For vehicle detection,a vehicle proposals frame algorithm based on point feature and edge feature is proposed.The image point information and the edge outline information may indicate the possibility that the image object exists.The point information can locate the target vehicle candidate frame but detect the non-object vehicle candidate frame,and the same edge feature can detect the accurate target vehicle candidate frame but a candidate frame with multiple objects and larger area will appear.The characteristics of performance differences and complementarity,the fusion of two features designed based on point points and edge proposals scoring method.Under the consistent evaluation criteria,the recall rate of the algorithm was as high as 0.9,and the accuracy rate increased by 10% to 15% year-on-year.