Abstract: Pedestrian video detection is increasingly important with the development of intelligent transportation. However, it is a challenging problem because pedestrians usually have unfixed shape, and are frequently influenced by several factors such as gesture, cloth, belongs, environment, etc. The pedestrian detection method presented in this paper is able to practice efficiently in actual traffic situation——a detection method based on codebook model in HSV color space. The codebook model is used to segment foreground and background in HSV color space. Then, the satisfied foreground is regarded as pedestrian with proper length-width ratio thresholds and area thresholds. Experiment results show that the presented method is effective and is able to be applied online.

Key words: intelligent transportation, pedestrian detection, video detection, motion detection

摘要： 近年来，随着智能交通的发展，利用视频对行人目标进行检测的意义日益重要. 但是由于行人是非刚体，且易受到姿态、服装、随身物品、外界环境等众多因素变化的影响，行人检测是一个非常具有挑战性的课题. 本文提出了一种可以从实际复杂交通运行环境中快速有效地检测行人目标的方法——基于HSV空间中密码本模型的行人视频检测方法. 文中首先在HSV色彩空间中应用密码本模型，有效地对前景运动目标和背景进行分割；然后，通过进一步对检测到的前景目标区域进行检验，将满足一定面积和长宽比阈值条件的前景区域判定为行人. 实验证明，本文提出的方法，具有较高的准确性，速度可以基本满足在线处理的需要.

关键词: 智能交通, 行人检测, 视频检测, 运动检测