本文为美国弗吉尼亚大学（作者：Douglas Grosvenor）的学士论文，共19页。

本论文的目标是设计一种算法来检测和跟踪静态视频中的人。现有的技术是不够的，取而代之的是一种基于背景减法的新方法。在使用静态摄像机时，该方法是成功的。在背景减法中，先计算出视频的背景，然后从视频的每一帧中减去背景。提出的两种算法都是基于受试者的重心，但第一种算法跟踪视频周围人的路径，使得视频非常混乱。这种实现对噪声的抵抗能力不强，因此开发了一种新的算法，通过在人周围放置一个绿色矩形来跟踪人体。最后，该项目成功地检测并跟踪了一个人在视频期间的行走，同时将其显示提供给用户。由于这只是一个概念上的证明，检测和跟踪步骤并没有实时运行，需要进一步改进算法。一种改进该算法的方法是将代码移植到CUDA，以利用图像处理中固有的并行性。

The goal of the thesis project was todevise an algorithm to detect and track people in a static video. Existingtechniques are inadequate; instead a new approach based on backgroundsubtraction is used. The approach is successful with a static camera. Inbackground subtraction, the background of the video is calculated a priori andthen subtracted from each frame of the video. This isolates the foreignobjects, which are detected via two simple algorithms. Both algorithms arebased on the subject’s center of mass, but the first algorithm traces the pathof the person around the video, making it very cluttered. This implementationwas not very resilient to noise so a new algorithm was developed which trackedthe person by placing a green rectangle around them. In the end, this projectsuccessfully detects and tracks a person walking for the duration of the videowhile displaying it to the user. Since this is only a proof of concept and thedetection and tracking did not run in real-time, further work is needed toimprove the algorithm. One suggested method to enhance this algorithm involvesporting the code over to CUDA to take advantage of the inherent parallelism inimage manipulation.

1. 项目背景

1.1 问题描述

1.2 图像分割

1.3 检测与活动轮廓

1.4 GPU与CUDA

2. 相关工作

3. 本文组织结构

4. 砖墙

5. 检测

6. 跟踪

更多精彩文章请关注公众号：