| Computer vision refers to the acquisition,processing,and analyzing of the image information of the surrounding environment through cameras,computers and other devices to realize the function of "seeing" similar to human vision.In the field of computer vision,the research of video-based multi-targets detection and tracking technology is a very meaningful and challenging task.The technology can be applied in many fields,such as intelligent monitoring,video indexing,human-computer interaction and the like.Here,intelligent monitoring means that without human participation and in terms of the digital images or videos taken by the camera,a computer can use technologies such as image processing algorithms and computer vision technologies to analyze and understand useful information and then to achieve the detection,identification and tracking,as well as behavioral analysis,of the moving targets.At present,when the multi-targets detection algorithm is applied in practice,due to the impact of light,the targets in the images taken from the cameras will have some shadows,which would lead to false positives or missed detection that affect the tracking effectiveness directly.Therefore,it is of great research significance to implement a stable and accurate detection algorithm in multi-targets detection and tracking system.The research and application of Multi-targets detection and tracking are challenging technologies.Up to now,there is no a universal detection and tracking algorithm suitable for all situations.It needs to choose or improve one or more algorithms for applications according to specific scenarios.This paper aims to study the detection and tracking of multi-robots and improve some existing algorithms.The main work of this paper is as follows:(1)This paper builds a multi-robots detection and tracking system platform according to the project requirements,and introduces the related hardware,software,and the overall design flow chart of the system.(2)Review the commonly-used moving target detection algorithms,including the interframe difference method,the optical flow method,the single Gaussian background modeling,the mixed Gaussian background modeling,and the Codebook algorithm.All of these algorithms are applied to the experimental platform to analyze their advantages and disadvantages such that the optimal detection algorithm among them,i.e.,the Codebook algorithm,can be selected.Although,the Codebook algorithm should be improved according to the actual need.The experiments show that the improved Codebook algorithm can effectively solve the problem of shadowing when detecting objects.(3)Due to the irregular shape and increased height of robots,based on the traditional detection algorithms,the projections of some irregular visual information of robots can be obtained.However,the projections have different errors in different positions and the overall error is usually large,which cannot meet the laboratory requirements for robot positioning.Therefore,this paper presents a detection algorithm based on local features for the robot platform,which greatly improves the accuracy of the position information of the objects.(4)The calibration algorithm is studied.According to the experimental environment,a suitable camera calibration method is selected.The method is of high precision and practicability,and provides a strong guarantee for providing high-precision robot position information.Combined with the detection algorithm in this paper,the error of position information of robot is calculated and analyzed.(5)Multi-moving objects tracking algorithm is also studied.According to the experimental needs,a Kalman filter-based tracking algorithm is applied.The mutual feedback mechanism between detecting and tracking is adopted to reduce the detection range according to the tracking result,which not only reduces the influence of interference during detection but also improves the detection accuracy.Moreover,the reduced detection range greatly reduces the detection time and improves the system's real-time. |
PDF Full Text Request