多摄像头目标检测与跟踪方法研究

华中科技大学

硕士学位论文

多摄像头目标检测与跟踪方法研究

姓名：颜杰

申请学位级别：硕士

专业：通信与信息系统

指导教师：邱锦波

2011-01-18

华中科技大学硕士学位论文

摘要

视频跟踪是计算机视觉领域的一个基础的研究课题，也一个非常具有挑战性的研究方向。在当前的现实生活中，视频跟踪技术已经在各种领域内得到了广泛的应用，其中包括视频监控、军事工程、交通管理、智能机器人和人机交互等，具有很高的学术研究和应用价值。

单摄像头的视频跟踪系统存在很多无法解决的问题，其中包括目标遮挡、摄像头视野有限、不能进行全方位的跟踪等问题，而多摄像头的跟踪系统能够很好的克服这些问题。因此，多摄像头目标检测与跟踪正在成为研究的热点。本文在前人研究的基础上，重点研究了如何提高多摄像头之间目标确认的精度，以及如何在保证对目标准确跟踪的条件下，降低整个系统的数据传输量和计算量。

本文首先分析多摄像头跟踪领域中，摄像头之间目标确认问题，提出了一种在基于平面单应性的确认技术中，引入目标距离特征的新方法。由于目标距离不受平面单应性约束条件的影响，加入目标距离特征能有效的提高摄像头之间目标的确认精度。实验结果表明，在基于平面单应性的确认算法中，增加目标距离特征后，确认精度得到了一定的提高。

为了有效地减少多摄像头跟踪系统的数据传输量和计算量，本文还提出了一种基于最优摄像头选择的跟踪算法，并从理论分析和实验上，对该算法的性能进行了评估。实验结果显示，该算法在不降低对目标跟踪准确度的情况下，有效地降低整个系统的数据传输量和计算量。

关键词：多摄像头，目标检测，目标跟踪，目标确认，最优摄像头选择

华中科技大学硕士学位论文

Abstract

The visual tracking is a basic research topic in the field of computer vision, but also a very challenging research direction. In the current real life, visual tracking technology has been widely applied in various fields, including video surveillance, military engineering, traffic management, intelligent robots and human-computer interaction and so on, and has high value of academic research and application.

Single-camera visual tracking system has many unresolved issues, including the target occluding, the limitation of camera field of vision, can not be omnibearing tracking and so on, but multi-camera tracking system can well overcome these problems. Therefore, Multi-camera object detection and tracking is becoming a hot research. Based on previous research, the paper focus on how to improve the accuracy of target identification between multi-camera, and how to reduce data transmission and computation in the overall system, ensuring the accurate target tracking.

First of all, the paper analyzes the target recognition problem between cameras in the field of multi-camera tracking, and puts forward a new method. The method introduces the feature of target distance into the homography-based technique of identification. As the plane homography constraints can not impact the distance between targets, adding the distance feature can effectively improve the target identification accuracy between different cameras. The Experimental results show that the new method can improve the accuracy of target identification, when the homography-based recognition algorithm is added the distance feature.

In order to effectively reduce data transmission and computation in the multi-camera tracking system, the paper also proposes a new tracking algorithm based on the optimal camera selection, and evaluates the performance of this algorithm from the theoretical analysis and experiment. The Experimental results show that the algorithm can effectively reduce the transmission and computation of the tracking system, but not loss of tracking accuracy.

Keyword: multi-camera，target detecting，target tracking，target identification，the optimal camera selection