A Research On Depth Sensor Based Human Tracking System

Human tracking is essential for service oriented robotic technology,however,traditional tracking algorithms are mainly based on monocular RGB camera,which has relatively weak perception of the environment compared to depth sensor,and due to the limitations on hardware and computational capacity of the robot,it is rather difficult to give consideration to both high accuracy and instantaneity.To solve the above problem,this paper will propose a highly efficient tracking system based on depth camera,the whole system should run on an IPC with high performance.This article is mainly about the tracking algorithm,the works that had been done can be listed as follows:1.Proposing a continuously adaptive template matching algorithm for human tracking in real time.The tracking window is adjustable based on the distance of the target,and template matching is used for further tracking.2.Implementing a mass center based adjustment algorithm for the tracking window,which solve the problem of window shifting with template matching.3.A novel approach for foreground separation in depth image with longer rectangle kernel,which eliminate the interference of noise from raw data.4.Implementing multiple helper strategies in GDB and utilizing Octave for developing the tracking system,simplify the debugging process related to machine vision.5.Using Valgrind tool for tracking process profiling,a good instantaneity is shown in experimental results.