Recognition Of Pedestrians’ Causal Relationship Based On Momentum Dynamics Model

Pedestrian behavior analysis is a hot issue in the field of machine vision. The existing researches are the identification and classification of the target behavior and abnormal behavior analysis. While the deep-seated theoretical basis which cause the target in the video behave is rarely studied.The relationships between the targets after years of academic studied and concluded, now are divided into three categories, namely, cause, prevent and enable. It have been well studied the identify issues related to the relationship between multiple targets by foreign scholars. It is, however, existing methods, including the force theory, mental model theory and cause model theory are objective studied in aspects of property and theory, and are difficult to apply in the analysis of the reality scene. Whether the force model or the force dynamics model, elements of which are virtual imagine based on reality, is difficult to make a quantitative calculation. In this paper, in order to study the pedestrians’ relations in actual scenes, we proposed an algorithm that is identification method among targets based on the momentum dynamics model. It can be used to identify and quantify the relationship between the targets. Specific methods are as follows:(1)First of all, tracking the targets with Cam-shift algorithm based on the color feature extracted from the selected targets so as to obtain and record the information of the targets’ location and time.(2)Build momentum dynamics model based on the concept of causality, combined with momentum theorem and social force model in the field of crowd behavior analysis. Reform the estimation method based on the momentum dynamics model.(3)Proposed the concept of causal value through the derivation of three existing relations, and based on causal determination method, dividing the causal value range, so as to recognize the relationship among multiple objectives intelligently.Experiments show that the proposed method can overcome that other models are not quantitative causal relationship between targets to achieve the goal of intelligent video surveillance more accurate identification of the target inter-relationships.