Research On Moving Target Detection Technology Based On RGB-D Data

The detection of moving objects is one of the hot issues in the field of computer vision.It is widely used in human-computer interaction,intelligent transportation,battlefield detection and other fields.This paper focuses on the problems of moving object detection based on color information or depth information,discussing from perspectives of super-resolution reconstruction of depth map,moving object detection based on RGB-D data fusion,pedestrian counting and software design.The main work is as follows:(1)In order to solve the problem of low resolution,noises and so on,a deep image super-resolution reconstruction method based on the second-order generalized total variational model is designed,which transforms the reconstruction problem into the optimal solution problem.First,the data items of the objective function are constructed according to the reconstructed constraints of the depth map.Then,the regularization model is weighted by the edge information of the RGB image,and the diffusion tensor is introduced into the second-order generalized variational model to form the regularization model for the depth image features.Finally,a high-resolution depth map is obtained by iterative weighting and original-dual algorithm.The experimental results show that this method can effectively solve the problem of low resolution of image depth,existence of holes and noise.(2)In order to solve the problem of moving object detection based on RGB or depth information respectively,a codebook algorithm fused with RGB-D data is proposed to detect the moving target.The depth information is used as the fourth dimension channel of the codebook to enhance the background Modeling,and foreground detection.The edge of the test result is clearly defined by image subtraction,mean filtering,threshold judgment,logical AND operation and so on.Experiments show that the algorithm overcomes the shortcoming that the color information is sensitive to light and shadows,or can not be detected by the depth information alone,and the edge noise is large.(3)On the basis of the moving object detection method designed in this paper,a pedestrian counting method is designed.The moving target detection algorithm based on RGB-D data fusion is used to detect the target area,and the target area is virtual counter.The motion block is divided and the virtual line is set to count and the pedestrian movement direction is judged.Using Kinect camera real-time collection of indoor control environment and complex environment based on two different groups of experimental video validation.(4)A set of moving target detection software was designed based on OpenCV library and Qt tool on VS2010 software platform.The depth map reconstruction algorithm,moving object detection algorithm and pedestrian counting method were integrated in the paper.The Kinect camera was used to collect and count in real time.The detection software integrates video input,pedestrian counting,comparison of detection and reconstruction results,virtual line calibration and so on,and it is convenient for the user to count pedestrians.At the same time,the feasibility of the algorithm is verified by the test system.