Foreground Detection Based On Mobile RGB-D Camera

With the progress of science and technology,the application of computer vision has been developing rapidly in many fields,and foreground detection,as the basis and premises to solve many computer vision problems,has been a research hotspot.Foreground detection is to extract the region of interest or moving target in the image or video as foreground from the background by certain technology.With the application of foreground detection becoming more and more extensive,the fixed camera can no longer satisfy all scenes.Due to the increasing popularity of mobile shooting equipment,the application of mobile camera in foreground detection is becoming more and more extensive.The Vibe algorithm is one of the foreground detection algorithms.It has many advantages such as small computation and good real-time performance.However,when the Vibe algorithm is applied to the scene captured by the mobile camera,some problems will occur,for example,the background model update mechanism cannot be adapted to the real-time scene.In order to solve these problems,this paper proposes an RGB-D Vibe algorithm,which improves the Vibe algorithm from three parts:background modeling,pixel classification and background model update.The difference between it and Vibe algorithm is that RGB-D Vibe algorithm also uses depth information.Depth information is obtained from depth images collected by RGB-D camera Kinect.The results of experiments show that the RGB-D Vibe algorithm can detect moving objects in a dynamic background effectively.In view of the foreground camouflage problem,this paper proposes a foreground detection method based on depth edge.This method uses depth edge detection to extract the foreground based on the assumption that the points in the external contour of the foreground are foreground points.Combining RGB-D Vibe algorithm with depth-edge-based foreground detection method,a foreground detection strategy combining RGB-D Vibe with depth edge extraction was proposed.This strategy uses color information,depth information,angle information and edge information to detect the foreground.A variety of information supplement each other,so that the foreground detection strategy combining RGB-D Vibe with depth edge extraction is suitable for more scenes.In this paper,Visual C++programming is adopted to realize the foreground detection strategy combining RGB-D Vibe with deep edge extraction.The strategy was verified experimentally with public dataset and self-generated dataset,and compared with RGB-D Vibe algorithm.The experimental results show that the foreground detection strategy combining RGB-D Vibe with deep edge extraction can effectively detect the foreground in the scene captured by the mobile camera,and compared with the RGB-D Vibe algorithm,the foreground detection strategy combining RGB-D Vibe with deep edge extraction has significantly improved the recall rate and other experimental indicators.