中文摘要：

为了在复杂交通场景分析中进行行为建模，传统的视觉单词通常只利用方向信息建立基础视觉单词，由于未对速度以及上下文状态进行特征提取，缺乏运动速度信息以及上下文信息，在行为分析过程中无法有效建模、鉴别运动目标运动轨迹细节信息相同但运动过程各异的行为。针对上述问题，提出了一种新时空视觉单词，该单词模型加入了较丰富的速度信息以及运动目标暂停上下文信息。实验结果表明，该视觉单词生成算法能有效提高交通行为分析性能。

英文摘要：

To model the complex traffic scenes in activity analysis, traditional visual words generally only employ the direction in- formation, which means it＇s not efficient in target tracking, behavior recognition and scene classification because it lacks speed and context information. In order to resolve the problem mentioned above, a new visual word model for the adaptive speed quantity process is proposed to establish the velocity word and solve the problem of lacking the velocity information, and Stop word concept to set up the moving target stop state. The experimental results show that this visual word generation algorithm can effectively improve the analytical performance of the traffic video.