基于多观察层次动态贝叶斯网络的视频行为分析

A multi-observation hierarchical dynamic Bayesian network for video behavior analysis

为了分析视频监控中行人和车辆的行为,本文提出了一种多观察层次动态贝叶斯网络模型。首先,行人和车辆的行为用静态、动态及相互关系特征来表示。然后将其作为模型的输入,通过网络提取和分析目标的行为及其相互关系。同时,本文设计了一个简单的模型选择准则,从候选模型池中选择适合当前场景的模型来减小计算复杂度。实验结果表明本文提出的方法能有效的分析视频监控场景中行人和车辆的行为。

A multi-observation hierarchical dynamic Bayesian network is present to analyze the behaviors of human and vehicle for video surveillance. First, the behaviors of human and vehicle are represented by static, active and related features. Then the model is built to detect and analyze behaviors and their relationships, which inputs are behavior representations. At the same time, in order to decrease the complexity, a simple model selection is designed. Experiments show that the proposed framework could efficiently analyze behaviors for video surveillance.