摘要目前的视频监控系统一般都会包含非常多的摄像头,每个摄像头一般都监控 一个固定的区域。在需要监控某一处的信息时,监控人员根据时间地点切换不同 视角来进行实时监控。通过结合行人检测,行为分析和视频监控等技术,可以为 实时的安全监控提供一种解决思路和方案。并且,对于安全领域而言,行人检测 可以很好的提供安全预警,避免发生安全事故。并且对于智能汽车领域来说,行 人检测系统可以实时检测周边的环境状况,检测周围是否有行人,以此提示驾驶 员注意安全,这样就可以减少人员的伤亡以及财产损失。80370
在本文中,主要研究了 HOG 特征(即梯度特征直方图)。从概念,特征提取 的算法以及应用做了详细的介绍。其次,本文还详细的介绍了几种目前实用性比 较强的行人检测分类算法:支持向量机(Support Vector Machine)、AdaBoost 算法。 了解并学习 SVM 算法的研究状况,并且以此做了相关的实验来验证对于包含行 人图像的检测。
根据实验结果我们可以得到一个结论:在只使用 HOG 特征来检测行人的情 况下,检测的准确率并不是特别高,而当将 HOG 特征结合 AdaBoost 分类算法 的时候可以有效的提高行人检测精度,降低误检率,提高行人检测的正确率,并 且在各种各样的自然场景下都有较好的识别率。 关键词:行人检测;多特征;AdaBoost 分类算法
Abstract The current video surveillance systems usually contain a lot of cameras, each camera generally monitor a fixed area。 When you need to get the information about a place, the monitoring personnel to switch different perspectives according to time and place to perform real-time monitoring。 By combining pedestrian detection, behavioral analysis and video surveillance technology, can provide a solution ideas and solutions for real-time security monitoring。 And, for the purposes of security, pedestrian detection can provide a good safety warning to avoid accidents。 And for intelligent automotive field, the pedestrian detection system can detect the surrounding environment in real time, to detect whether there are pedestrians around, in order to prompt the driver to pay attention to safety, so that you can reduce casualties and property losses of personnel。
In this paper, the main characteristics of HOG (gradient feature histogram)。 From concept algorithm, feature extraction and application of a detailed introduction。 Secondly, this paper also introduces a common classification of pedestrian detection algorithm: support vector machine (Support Vector Machine)。 Learning to understand and status SVM algorithm, and thus do the relevant experiments。
According to the results we can get a conclusion。 HOG features combined with SVM classification algorithm can effectively improve the pedestrian detection accuracy and reduce false detection rate, which can improve the accuracy of pedestrian detection, and has a good recognition rate in a variety of natural scenes。
Keywords:pedestrian detection, multi-feature, AdaBoost sort algorithm
目 录
第 1 章 绪论 5
1。1 研究背景以及研究意义 5
1。2。1 行人检测现状 6
1。2。2 行人检测存在的难点 6
1。3 本文研究的内容和组织结构 7
第 2 章 对人体的特征描述研究 8
2。1 引言 8
2。2 Haar-Like 特征 8