微空间场景与视频分析相结合的审讯室异常行为检测

行为检测是智能视频分析的研究热点。目前,行为分析主要是基于视频图像空间,而丰富的空间场景信息并未得到有效利用。事实上,个体行为与空间场景紧密关联,不同场景可能具有不同的行为表征。本文以审讯室的典型微空间场景为例,以刑讯逼供行为检测为研究内容,探究空间场景约束下的个体异常行为检测方法。首先,从行为变量的概念出发,解析了刑讯逼供的行为变量和行为类型;其次,分析了刑讯逼供行为变量的视频特征,并设计了相应的行为变量检测模型;再次,通过灭点计算实现了空间场景和视频图像的双向映射关系的解算,并在此基础上设计了顾及空间场景的刑讯逼供行为分级分类检测策略;最后,通过模拟实验对本文方法的可行性进行了分析验证。

A Method of Abnormal Behavior Detection in Interrogation Room Based on Video Analysis Combined with Micro-spatial Environment

The detection of anomalous human behaviors has received tremendous attention in the research of intelligent video analysis. However, most of existing methods for anomaly analysis are based on image space, and abundant information in the geographical space goes unused. In fact, human activities are closely related to geographical spaces, and different scenario types may correspond to different classes of human behaviors. So, in this paper we chose interrogation room as a representative of the micro-spatial environment and conducted a study on the anomaly detection of extorting confessions by torture, considering the spatial constraints. Firstly, the concept of behavioral variables has been stated, and then behavioral variables and their classes of extorting confessions by torture have been deeply analyzed. Secondly, the video characters of behavioral variables of extorting confessions by torture have been discussed and their corresponding detection models have further been designed. Plus,a bi-directional mapping model between the geographical space and the image space has been constructed according to vanishing points. This proposed mapping method can effectively avoid the calculation of intrinsic and extrinsic camera parameters. Based on above, a hierarchical and classified strategy for anomaly detection of extorting confessions by torture has been developed, considering the spatial constraints to human behaviors without learning and training processes. At last, the presented method in this paper is tested and verified by a series of simulated experiments.