Research And Realization On Interested Moving Object Detection

Target detection is one of the most important problems in image and video analysis,pattern recognition and computer vision applications.It plays an important role in various fields such as video surveillance,vehicle navigation,robot vision and intelligent transportation system.In general,when viewing a video,people usually tend to be interested in a specific moving target called interested objects.Accurate and fast extraction of the interested object will greatly improve the effectiveness of follow-up and recognition processing.In practical applications,facing the massive video images and different applications,it is urgent to solve the problems about accuracy,real-time and platform versatility of interested moving object detection.Specific research work is as follows:Considering that moving objects in video are all the interested objects,then the image segmentation based on energy function is one of the most widely used methods.Aiming at the problems of the low computing efficiency and platform limitation on image segmentation algorithm based on energy function theory,an OpenCL-based parallel implementation of Continuous max-flow algorithm is proposed in this paper.The optimization problem of iterative solving max-flow is parallelly implemented based on parallel characteristics of algorithm analysis.High efficiency of algorithm as well as platform transition is achieved on different hardware architectures by reasonably invoking GPU and CPU.Then under the mixed Gaussian background model,the proposed parallel implementation of Continuous max-flow algorithm is used to detect the interested objects in video.The experimental results show that the parallel algorithm invoking GPU and CPU achieved a significant improvement of computing speed compared with the CPU serial algorithm on the premise of ensuring the quality of video image segmentation;The proposed algorithm run successfully on three main platform including AMD,Nvidia and Intel,which effectively verifies the validity and the cross-platform ability.And it basically satisfies the requirements of practical application.People are not interested in all the moving objects in actual video analysis,so it is necessary to extract a certain kind of specific moving objects according to different applications.Aiming at the problem that the moving target can not be detected accurately due to the interference of complex background and other non-interested moving objects,this paper proposes a new detection framework for interested moving objects.On the basis of the traditional markov random field(MRF),the region of interest suggested by the haar-like cascade classifier is introduced as the high-order potential energy term,then high order MRF model is constructed.In this unified energyframework,the problem is transformed to the energy optimization problem which can be solved by max-flow/min-cut.The experimental results show that the expression ability of the model is improved by introducing regions-of-interest as high-order potential energy term,which is obtained by cascade classifier based on haar-like feature.At the same time,the algorithm effectively enhances the segmentation effect of the interested objects,making the interested objects have a better edge effect,improving the accuracy of segmentation as well as the visual effect of interested objects.