Research Based On Image Sequences For Moving Object Detection And Tracking

Target detection and tracking technology is a focus research in the field of computer vision,which is widely used in people's daily life and work.With the rapid development of artificial intelligence,a large number of image sequences are generated,so the needs for target detection and tracking are increasing.However,there are a lot of external environmental factors in the real scene,such as sudden change of illumination,weather changes,Character shadows,and so on,which make there is no perfect method for a variety of complex scenarios so far.Therefore,detection and tracking of moving objects is still a great challenge in the field of computer vision.The main results of my paper are as follows:First,moving object detection based on improved Spectral Clustering,the main work is as follows:1.Aiming at the problem of the spectral clustering algorithm in finding the weight matrix,which can not reflect the image space information by using the weight matrix obtained by Gaussian function.In this paper,we introduce a new function containing the elongation factor,which takes spatial information of the moving feature points into account,which makes the weight matrix more accurate;2.The motion feature points obtained by the background subtraction method can well reflect the motion characteristics,but it also contains too much noise.In my paper,the clustering method is used to classify the feature points so as to separate the background from the foreground.Compared with the threshold selection,the results using clustering method are more accurate.3.Based on the feature points obtained by background subtraction method,my paper uses improved spectral clustering algorithm to cluster.It show that the proposed algorithm greatly improves the performance of target detection through simulation experiments.Second,moving target detection based on optical flow field and EM algorithm,the following work has been done:1.According to a large number of experiments and data analysis,the optical flow field can retain the motion characteristics of each pixel well in the target detection,but there contains too much noise,which make the target detection effect is not accurate,So we use clustering method to filter the optical flow and eliminate noise;2.EM algorithm is a soft clustering method based on a posteriori probability.After using two pictures to obtain the optical flow field,on the basis of the optical flow field,we use EM algorithm to cluster the optical flows belonging to the background and target motion field,respectively,which make extraction of moving target more clearly.Third,based on the improved Vibe algorithm for moving target detection,the following work has been done:1.Aiming at the phenomenon that the Vibe algorithm has a ghost image in the target detection process due to the conservative updating strategy,In my paper,we introduce a learning rate,after the background model is initialized and the target detection is completed,the data set composed of the current pixels and the data set composed of a random background model are added to the background model according to the learning rate,which makes the background model more abundant.2.In my paper,the information of the foreground moving object is added in the new background update strategy when the background model updating.So randomly select pixels from the interlaced neighborhoods to fill the moving target region of the current frame and then added to the background model which make the model is closer to the real scene;3.The Vibe algorithm with learning rate is used to perform target detection,it is modified by the open operation of morphology,which makes the target detection more accurate.