Study On The Algorithms Of Moving Objects Detection In The Video Sequence

In the computer vision field, foreground detection is to separate the moving objects of interest from the background as an important pre-processing task. It is now widely applied in the automatic video survalliance (AVS), video compression, video index and automatic human-machine interface. In the realistic scenarios, how to accurately separate the foreground from the background is very challenging because of illumination change, background elements motion, shadow or camera motion.We focus on the foreground detection methods in the dynamic scenes. One scenes are shot by the static camera but with the scene motion (such as fourtain, waves, flying snow and so on). Another scenes are static but with the camera motion. Even some scenes are more complex, where both scenes motion and camera motion exist. Any of these types of scenes are refered to as the dynamic scenes. We mainly study on Mixture of Gaussian (MOG), Dynamic Texture (DT) model and some methods based on center-surround framework.In order to reduce the computational cost in the relatively static scenes shot by the stationary camera, the MOG method is improved. The improved method first detects the rough fouregrund region by Running Average (RA) algorithm, where each pixel is then processed by edited MOG algorithm. In order to suppress the shadow, the YUV color information is used as the pixel feature. Compared to MOG and No-Parametric Kernel Density Estimator (KDE) algorithm, the improved method has the bettter performance and lower computational cost. Its processing speed can meet the realtime need.We do much research on the DT model for some dynamic scenes shot by the stationary camera.When the video process is modeled by DT in a holistic manner, the observed data matrix has the high dimension. So Singular Value Decomposition (SVD) of the the observed data matrix has high computational complexity. In order to resolve this problem, the Sustain Observibility (SO) method is modified. It first improves the method of measuring the observibility and then measures the observibility at the subsample locations according to the system property of observibility. The observibility value of each pixel at the original scale can be obtained by upsample technique. The modified method has the similar performance as the SO method, but its computational cost is much lower. In order to reduce the dimensionality of the model, the DT model is applied to local video patches, which reduces the computational complexity of each SVD operation, but the number of DT model is increased. In order to deal with the problem, we propose the Local Dynamic Texture (LDT) method, which computes the similarity between the video patches according to the dynamic redundancy. DT model is only applied to the video patches with little similarity. Compared to other methods, LDT method both has the lowest average Equal Error Rate (EER) values and computational cost.Most approaches based on DT model assume that the dimensionality of the state space is a constant for all the tested scenes. In order to deal with this problem, we propose an adaptive method which is driven by the observed data, which computes the singular entropy from the singular matrix. The increment of singular entropy at each order is thresholded to decide the order of model. When the dimensionality of the state space is adaptively decided according to the proposed method, the lowest EER can be obtained by applying DT model to foreground detection.Besides, in order to avoid the SVD operation, a method of combining the batch-PCA(Principal Component Analysis) and Candid Covariance-Free Incremental Principal Component Analysis (CCIPCA) is adopted to reduce the computational cost, which the basis parameters are learned by batch-PCA, and other DT parameters are obtained by updating the basis parameters with the new observed data. Compared to batch-PCA, the method has the similar performance, but its computational cost is much reduced.Most approaches detect some background as the foreground in the scenes with camera or Ego motion. In order to resolve the problem, much research is done on the center-surround hypothesis from the biologic vision and we propose a method which combines the global and local detection. In the stage of global detection, the improved SO method is used to obtain the candidate foreground regions, In the stage of local detection, bayesian center-surround framework is used to compute the local feature contrast in the the candidate foreground regions. The contour information obtained from the local detection is feedback to confirm the accurate foreground regions and remove the background pixels in in the candidate foreground regions.The global detection is operated at each pixel of the whole video frames with the low algorithm complexity, while the local detection is only done at at pixels in the candidate foreground regions with the high algorithm complexity. So the average processing time per frame is reduced greatly. Compared to other methods, the proposed method has the better performance and lower computational cost.