基于视频图像的交通事件自动检测算法综述

摘要

Firstly, this paper analyzed research progress of traffic automatic incident detection, and depicting object detection, object tracking and behavior recognition and understanding. Obtained low level information from object detection and tracking, traffic automatic incident detection needed deep level understanding and recognition for tracking result. Secondly,introduced HMM( hidden Markov model )and SOFM( self-organized feature maps) importantly in motion understanding and behavior recognition. Finally, analyzed technical difficulties and their possible solution from motion segmentation and feature extraction aspect, as well as forecasting the possible research direction.%首先分析了交通事件自动检测的研究现状,对其中涉及的目标检测、目标跟踪以及行为识别与理解进行描述.目标检测和跟踪得到的是底层信息,而实现交通事件的自动检测需对跟踪结果进行更深层次的理解和识别.然后重点介绍了运动理解和行为识别中的HMM(隐马尔可夫模型)方法和SOFM(自组织特征映射神经网络)方法.最后从运动分割和特征提取方面分析了技术难点及解决方案,对可能的研究方向进行一定的预测.