Moving Object Detection Based On Color And Depth Information

Moving object detection is to extract the moving object of interest from the video sequence by certain detecting method.It is a hotspot in the field of computer vision,video surveillance,pedestrian detection,artificial intelligence.And as the foundation of image sequence analysis process,the test results of moving object detection have an important impact on subsequent processing.In recent years,numerous scholars at home and abroad have proposed many methods for different scenarios.Most of these methods are based on color information and get better results to some extent,but all cannot solve the problem of color camouflage.The use of depth data provides a strong basis for solving this problem,but the problem of depth camouflage will occur only with depth data.In this paper,based on the features of the color map and depth map obtained by the Kinect camera,the different moving object detection methods are used respectively.Then we combine the binary images obtained from the two methods effectively,and the results are better than the two algorithms used separately.Firstly,the methods of moving object detection based on color data are numerous,and the methods that only involves the feature of pixels are sensitive to illumination.Therefore,this paper proposes a moving object detection algorithm based on the local binary similarity patterns(LOBSTER algorithm)in the color image.The algorithm uses the LBSP descriptor with spatial information to construct the background model.Tests on CDnet2014 datasets show that the algorithm is better than most algorithms.Secondly,there are some pixels without value and a large number of noise points in the depth image obtained by the Kinect camera.We propose to construct models for pixels with depth values and without depth values separately to generate a hybrid model.As the initial background model,the hybrid model is added to the object detection algorithm of mixed Gaussian background model.The method can reduce the error of object detection and reduce the influence of noise.The binary images of the color map and the depth map are obtained by the use of the first two methods.The two binary images are combined by logic or operation to get the rough moving object,but the edge of the moving object may be irregular or partially missing.For this problem,we use Canny edge detection algorithm for differential calculation according to background difference method principle and get a more complete profile of moving object.Then,we combine the edge detection difference map and the previous binary image by logic or operation.At last,the morphological processing operations in the filling area and the median filtering operation are used to obtain a more complete moving object.At the same time,it can also suppress the noise.