基于粗糙集的多类人群异常行为识别算法

为解决现有基于人工设计特征行为识别方法缺少多类异常行为分类研究和受人工影响大等问题,提出和实现了基于粗糙集的多类中低密度人群异常行为识别算法.该算法首先提取目标人群的人数、帧平均加速度、矩形框的距离势能、方向混乱熵,以及帧间混乱程度五个运动特征量,利用粗糙集从中学习以获取决策规则,再对正常、四散、同向加速跑、突然聚集和群殴这五类人群行为进行分类,并定量对比分析本文算法和其他同类算法处理同一视频集的分类效果.结果表明:与随机森林法等其他同类算法相比,该算法不仅能够有效检测出人群异常行为,还能准确地对五类人群行为进行分类,其识别准确率和覆盖率均有明显提升.

An Algorithm for Identifying Multi-class Abnormal Behavior of Population Based on Rough Set Model

In order to solve the problem that existing artificial design feature behavior recognition method lacks multi-class abnormal behavior classification research and is greatly influenced by manpower, a multi-class middle-low density crowd abnormal behavior recognition algorithm based on rough set was proposed and implemented. The algorithm firstly extracts five motion characteristics of target population, average acceleration of the frame, distance potential energy of the rectangular frame, entropy of direction chaos and degree of inter-frame chaos. Using rough set to acquire decision rules, then classified the behavior of normal, scattered, accelerated run, sudden clustering and group fight, compared and analyzed on classification effect of the same video set between this algorithm and other algorithms. The results show that compared with other algorithms such as stochastic forest method, this algorithm can not only effectively detect the abnormal behavior of the population, but also accurately classify five types of crowd behavior. Both recognition accuracy and coverage rate are improved significantly.