Research And Application Of Motion Shadow Detection In Video Scenes

Computer vision and image processing have a wide range of applications,and moving shadow detection,as a basic work of advanced computer vision and image processing,the accuracy of its detection results will directly affect the development of follow-up work.The moving shadow and the moving target have two similar characteristics: one is that the moving shadow and the moving target are obviously different from the background;the other is that the moving shadow and the moving target have the same moving characteristics.The above two characteristics can easily lead to the adhesion of the moving target,the distortion of the moving target and even the loss of the moving target,thus affecting the detection,tracking and classification of the target.Because of the similar characteristics between moving shadow and moving target,moving shadow detection is still a very challenging task.In this thesis,a moving shadow detection method based on semi-supervised learning is proposed to effectively distinguish shadow points from the foreground,and on this basis,a moving shadow detection system is designed and implemented.The research work of this thesis is as follows:(1)A moving cast shadow detection method based on semi-supervised extreme learning machine is proposed.The moving shadow detection method based on semi-supervised learning machine in this thesis mainly includes the following steps: firstly,the foreground image of video image is extracted by background difference method,and the foreground image including moving object and moving shadow is obtained;secondly,according to the physical model of shadow,pixel level features and region level features are extracted from the foreground.Then,semi-supervised extreme learning machine is trained by using the features extracted in the previous step,and the object and shadow are classified.In order to ensure the accuracy of the results,the classification results are post-processed.Finally,this method is compared with other excellent shadow detection methods in two common data sets,the results show that the comprehensive performance of the proposed moving cast shadow detection method based on semi-supervised extreme learning machine is better than that of several representative methods.(2)A moving cast shadow detection system is designed and implement.In this thesis,a moving shadow detection system in video scene is designed and implement,the requirements of moving shadow detection system are analyzed and its function is realized.First of all,the feasibility analysis and requirement analysis of the system are carried out,and then the summary design and detailed design of the system are carried out.Finally,the implementation and function test of the system are carried out.