一种基于改进粒子滤波的多目标检测与跟踪方法

摘要

针对复杂监控环境下由于目标间相互遮挡导致多行人目标检测与跟踪准确率低的问题，利用一种基于改进运动目标的检测方法和改进粒子滤波的多目标跟踪算法。首先，采用改进动态高斯模型在监控环境下对人体运动目标进行检测，然后利用改进粒子滤波算法对场景中多人体运动目标进行识别。该算法可以在发生行人目标发生遮挡时，消除或降低由于遮挡等原因造成的目标检测和跟踪精度下降问题。实验证明该算法在满足实时性的要求下，可以在遮挡和非遮挡情况下对人体目标进行准确检测和跟踪。%To solve the occlusion problem in multiple pedestrians detection and tracking with complex surveil-lance environment, the improved Gaussian detection model and improved particle filter were presented. The im-proved Gaussian detection model was firstly used to detect multiple moving targets under surveillance environ-ment. Then, the improved particle filter was employed for multiple pedestrian targets tracking while the targets were occluded. This method could eliminate or reduce filter instability and improve the tracking accuracy. Ex-periments proved that when the algorithm satisfied the real-time requirements, it could track human target accu-rately under conditions of both occlusion and non-occlusion.