复杂运动情况下的多目标检测与追踪

Multi-targets Detection and Tracking Algorithm under Complicated Movements

针对多运动目标视频序列中目标存在过缓运动且相互间遮挡的问题,提出一种改进的混合高斯模型(Gaussian Mixture Model,GMM)进行目标检测并结合卡尔曼滤波进行跟踪的新算法;在运动背景检测中,采用三帧差法和混合高斯模型的融合算法进行目标提取,解决了过缓运动目标检测区域不完整的问题;在运动目标跟踪中,由于Kalman滤波器采用目标检测结果进行预测,对观测噪声矩阵进行自适应更新,使得跟踪的稳定性得到加强;通过对比,验证了新算法对存在过缓以及遮挡的不规律运动情况的多目标检测识别与跟踪的实时有效性与准确性。

Owing to the problem of too slow movements and inter--block may come up in video sequences, a new algorithm is presented based on improved Gaussian mixture model and Kalman filter. In the background motion detection, a fusion of three frame difference method and Gaussian mixture model is proposed to solve the problem that slowly moving target detection area is incomplete. In the trace, Kalman filter use the target detection results to complete forecast tracking and the adaptively observation noise matrix to reinforce the stability of the tracking. Experimental results show that the algorithm can well ensure the detection and tracking of the slowly, blocky and irregular Multi --targets by comparing the previous algorithms.