一种基于光流和二级聚类的移动背景下的目标检测算法

针对现有的移动背景下的目标检测算法存在检测速度较慢、自适应性差和检测准确度不高的问题,提出了一种基于光流和二级聚类的移动背景下的目标检测算法;该算法融合了阈值自适应规则和基于优化检测结果的反馈机制。首先采用Lucas-Kanade光流跟踪算法和DBSCAN聚类算法提取出前景目标,然后采用改进的凝聚层次聚类算法将前景目标分类。在第一级聚类时建立基于初始聚类结果的自适应规则,实现了自适应地检测目标;在第二级聚类后,通过去除错误匹配特征点和阴影区域特征点优化检测结果;并将优化后的检测结果反馈给第一级聚类过程以更新适用阈值,使目标检测更准确。在多个视频库上进行验证,实验结果证明该算法检测速度快、自适应性良好、检测准确度高。

A Moving Objects Detection Algorithm In Dynamic Scene Based On Optical Flow And Secondary Clustering

There are several problems that slow detection speed,bad self adaptab- ility and high misjudge rate in existing moving objects detection methods in dynamic scene, so this paper proposed an new algorithm based on optical flow and secondary clustering ,which combined the adaptive threshold and feedback mechanism based on optimizing the detection results. Firstly, Lucas-Kanade optical flow tracking algorithm and DBSCAN clustering algorithm were used to extract the objects in foreground, then we used an improved agglomerate hierarchical clusteringSalgorithm to classify the objects in foreground. When the first-level clustering process run, it adjusted the threshold according to the clustering results to obtain applicable threshold , thus it could detect targets adaptively.The result of the second-level clustering was denoised and optimized and then it was fed back to the first-level clustering process, thus the applicable threshold would be updated to get more accurate detection results.This proposed algorithm was validated on several video library and the experimental results demonstrated its fast detection speed,good self adaptability and low misjudge rate.