MB-LBP特征提取和粒子滤波相结合的运动目标检测与跟踪算法研究

Research on Algorithm of Moving Target Detection and Tracking Based on MB-LBP Feature Extraction and Particle Filter

在复杂环境下,由于行人密度大以及运动随机性,导致运动目标(行人)难以检测和跟踪,造成人员计数误差。提出一种MB-LBP(Multi-scale Block Local Binary Pattern)特征提取和粒子滤波相结合的运动目标检测与跟踪算法来解决此问题。该算法首先用AdaBoost提取MB-LBP特征训练生成分类器进行人头检测,并根据人头目标尺寸变化范围去除部分误检,然后用改进的粒子滤波算法预测跟踪多个运动目标,最后对跟踪的运动目标进行计数。实验结果表明,提出的算法能够对复杂环境下多个运动目标进行有效检测及跟踪,准确、快速地对视频帧中的人员进行计数。

In complex environments, because of high density and movement randomness of pedestrians, it is too difficult to detect and track moving targets, therefore leading to counting errors. An algorithm of detecting and tracking moving targets combining the MPrLBP（Multi-scale Block Local Binary Pattern） feature extraction and particle filter algorithm was proposed in this paper. Firstly, we adopted the algorithm of AdaBoost to extract the MB-LBP features which are trained to generate a classifier, detect head targets, and remove part of false detection based on the size ranges of head targets. Secondly, we improved the original particle filter algorithm to predict and track moving targets. Finally, we counted the moving targets which are tracked. The experiments show that the algorithm can effectively detect and track multiple moving targets in complex environments, and count pedestrians in video frames accurately and rapidly.