Research On Adaptive Detection And Recognition Method Of Dynamic Video Image Target

Detection and recognition technology of video sequence images is an important part of intelligent video monitoring system in dynamic environment.It is also frequently used in the fields of image processing,computer vision and pattern recognition.Moreover,it has important research value in military guidance,security monitoring,intelligent transportation and other fields.The image of moving target is unstable and unclear due to the diversity of the environment and the random uncertainty of the distance between the target and the optical imaging system in the practical application environment,which limits the performance of scene monitoring system.Therefore,this study takes pedestrians as the research object,and studies the adaptive detection algorithm and multi-feature fusion recognition algorithm based on image processing with the help of the monitoring system composed of camera and cradle head.Then,this study has important theoretical significance and practical value.The results of this study as follows:Firstly,the method of camera automatic focusing and the image sharpness evaluation algorithm are introduced.In this study,an adaptive detection method combining Vibe background modeling and scale-invariant features is adopted to complete the adaptive detection of the target and extract the texture features and gradient direction features of the detection area of the target.Experimental results show that the adaptive detection algorithm can accurately and completely detect the moving target region in the process of camera automatic focusing.Secondly,a multi-feature fusion recognition method is studied to solve the effect of the scale and illumination intensity of dynamic targets on the recognition rate of a single feature.The structure and framework of multi-feature information fusion and recognition are designed and presented.The target is preliminarily identified and preprocessed by radial basis neural network,and the data are fused and identified by D-S evidence theory.The experimental results show that the recognition accuracy of multi-feature fusion is up to 95.00%,and the recognition effect is obviously better than that of single feature classifier.Finally,nuclear related filter tracking algorithm is used to improve the stability and robustness of tracking algorithm by texture feature and direction gradient feature fusion.Nuclear filter models are respectively trained by the extraction of target feature.The target location is confirmed by using the method of weighting to fuse output response diagram,which can achieve the target tracking.Experiment results show that the multi-feature fusion tracking algorithm has high tracking accuracy and stable tracking performance when the target and the background are similar in color,the target is disturbed by other objects,and the target is partially occluded.