| With the rapid development of multimedia technology and the reduction of monitoring equipment costs,road surveillance video is widely used in road safety management.Video keyframe extraction technology is considered a fast and efficient method for massive road surveillance video.For the road vehicle monitoring video,the currently used video key frame extraction method is not suitable for use,and the use of visual saliency is not sufficient..The purpose of this thesis is to extract key frames which can represent the original video content and make full use of visual saliency.The main work includes two aspects:(1)Aiming at the target vehicle in the road surveillance video,a keyframe extraction method based on multiple static feature fusion is proposed.The main work can be seen as follows: First,On the basis of vehicle tracking and target extraction,the underlying feature saliency map of the target vehicle and the license plate saliency map of the target vehicle are integrated,and the fusion of various static features better expresses the vehicle information.Second,Because this paper optimizes the low-level features according to the principle that the smaller the similarity of license plates,the greater the penalty for the underlying features.Therefore,the extracted key frames of the target vehicle can provide more effective samples in license plate recognition.Experimental results show that,for a series of vehicle images in the surveillance video,the algorithm presented in this paper effectively selects the most abundant image of vehicle,realizing effective compression of vehicle data for road vehicle surveillance.(2)Aiming at the traffic abnormal behavior of vehicles in road surveillance video,a keyframe extraction method combining dynamic and static saliency based on vehicle tracking and occlusion analysis is proposed.The specific work is as follows: First,the road surveillance video is segmented by the frame motion amount;Then,the moving direction saliency,the motion intensity saliency and the complete saliency of the moving vehicle based on the motion change and the occlusion of the moving vehicle are extracted.next,the visual saliency of the frame image is obtained by weighted fusion with fixed coefficients;Finally,the frame with the greatest visual saliency in the segment video is extracted as the current key frame.The segmented video sequence frame is arranged in descending order of visual saliency,and the frame larger than the segmented video average visual saliency is used as the candidate frame,and the absolute max value of the correlation coefficient between the candidate frame and the current key frame is compared with the threshold value todetermine whether the candidate frame is the key frame.If it is judged to be a key frame,the candidate frame is added to the set of current keyframe.The experimental results indicate,the keyframes extracted by this algorithm are not only representative of abnormal behavior frames with low redundancy,but also the moving vehicles are relatively complete,which preserves the inherent information of the moving vehicles. |
