利用鱼眼相机对人群进行运动估计

发布时间：2018-07-11 17:38

本文选题：鱼眼相机 + 特征点　；参考：《武汉大学学报(信息科学版)》2017年04期

【摘要】：人群运动估计是人群行为分析的重要步骤。特定场景的人群运动分析和监控,是维护公共安全和社会稳定的一个必要措施,也是视频监控领域的一个研究难点。利用鱼眼相机视场大、无视觉盲区的优点,提出了一种基于特征点光流的人群运动估计方法。首先,采用一种基于面积反馈机制的混合高斯背景差分方法,对原始视频图像进行预处理,并利用圆拟合的方法获取兴趣区域;其次,为了在保证准确描述人群目标的同时提高算法的实时性,提出一种基于边缘密度非均匀采样的人群特征点提取方法来描述运动的人群目标,并利用LucasKanade光流法计算光流场;最后,为了解决远近人群的尺寸大小不一致的问题和鱼眼相机的畸变问题,采用鱼眼相机的透视加权模型,计算人群运动加权统计直方图,获取人群在鱼眼图像中的全局运动方向和速度。实验结果表明,针对密集的人群,该方法能有效、实时地估计人群的运动方向和速度,为人群行为分析提供有力的研究基础。
[Abstract]:Crowd motion estimation is an important step in crowd behavior analysis. Crowd motion analysis and monitoring in specific scenes is an essential measure to maintain public safety and social stability, and is also a research difficulty in video surveillance field. Based on the advantages of large field of view and no visual blind area, a method of crowd motion estimation based on feature point light flow is proposed. Firstly, a hybrid Gao Si background difference method based on area feedback mechanism is used to preprocess the original video image and obtain the region of interest by circle fitting. In order to ensure accurate description of crowd targets and improve the real-time performance of the algorithm, a crowd feature point extraction method based on non-uniform sampling of edge density is proposed to describe moving crowd targets, and Lucas Kanade optical flow method is used to calculate optical flow field. In order to solve the problem of inconsistent size and distortion of fish-eye camera, the weighted statistical histogram of crowd motion was calculated by using the perspective weighting model of fish-eye camera. The global motion direction and velocity of the crowd in the fish-eye image were obtained. The experimental results show that this method can estimate the movement direction and velocity of the crowd effectively and in real time and provide a powerful basis for the analysis of crowd behavior.
【作者单位】：湖北大学计算机与信息工程学院;武汉大学电子信息学院;
【基金】：湖北省大学生创新训练项目(201510512041)~~
【分类号】：TP391.41

【参考文献】