Research And Implementation Of Algorithms For Object Detection And Shadow Removal In Visual Surveillance

Development of the computer vision technology, the electronic and the communication technology, has made the intelligent visual surveillance system an increasingly important safe defense way. Both its history and current situation is summarized here, then, the algorithms for the moving object detection and the shadow removal are all discussed.As the initial stage in the visual data processing, moving object detection is a key point, also, a difficult research area. After carefully study of moving object detection methods used presently, a more reliable algorithm is determined through experiments, that is, the background difference method based on mixed Gauss model. It has been proved in experiments that the mixed Gauss background model is robust.As shadow exists in the moving target detection results, in this paper, based on Gaussian mixture background model, we propose two kinds of method of the shadow suppression for pedestrians and vehicles. One is based on the reflectance and color constancy algorithm for removing the shadow of the pedestrians, this method hold that the neighborhood illumination of pixels is invariant and the background where is covered by shadow does not change its reflectance and color composition, we use this point to detect and remove shadow. The other is based on intrinsic image decomposition and wavelet transform for removing the shadow of the vehicles. First of all, the current image and background image which is created by Gaussian mixture are transformed into log-domain. Then obtaining the gradient maps of two images, and getting the gradient map of foreground image using the difference between the gradient maps of the two images, removing the shadow of the target area roughly using intrinsic image decomposition method and we get the edge map of foreground image. Finally getting the horizontal and vertical projection of the edge map, and obtaining wavelet transform of the projection, Find the vehicle area after removed shadow accurately. Finally, existing and new proposed shadow detection algorithm results were compared. Experiments show that the proposed shadow suppression algorithms of pedestrian and vehicle can accurately remove the shadow, with strong robustness.