一种鲁棒高效的视频运动目标检测与跟踪算法

提出了一种视频运动目标的快速检测和稳定跟踪算法. 目标检测使用减背景法, 用均值法构造背景图像, 提出一种基于熵能和广义高斯分布的局部自适应阈值选取算法, 可有效克服噪声的影响. 采用基于特征匹配的目标跟踪方法, 提出一种LICS (Logarithm illuminance contrast statistic)特征, 该特征能够更加充分有效地表征目标, 可在光照和目标姿态变化的情况下实现刚体目标的稳定跟踪. 使用Kalman滤波限制搜索匹配范围以减小计算量. 用目标子区域匹配的方法解决目标相互遮挡时的跟踪问题. 实验结果表明, 该算法在运动目标检测效果、跟踪稳定性和运行时间方面都有良好的性能.

A Robust and Efficient Video Moving Object Detection and Tracking Algorithm

A simple and efficient moving object detection and tracking algorithm is proposed. The object detection is based on the background subtraction method; an adaptive local threshold selection method on the use of entropy power and GGD (Generalized Gaussian distribution) is proposed to get over the noise influence. Feature based tracking method is used in object tracking. A feature named LICS (Logarithm illuminance contrast statistic) is proposed, which can effectively represent the objects' appearance. Tracking of rigid objects by LICS is stable when the objects' illumination and posture are variable. The Kalman filter is used to restrict the search window and reduce the calculation. A sub-block matching algorithm is used to handle the objects occlusion. The experimental results show that the proposed algorithm has good performance.