多信息融合的人体目标跟踪算法

Multi-information fusion of human target tracking algorithm

为了在智能监控系统中自动跟踪人体的运动轨迹,提出了一种融合颜色、深度和预测信息的人体目标跟踪算法,首先将颜色信息和深度信息融合实现该算法的目标模型,在目标跟踪时,将目标模型融合到Camshift跟踪算法中,然后通过卡尔曼滤波器对搜索窗口进行预测,最后在与之对应的深度图像中计算目标的质心位置,采用微软Kinect传感器,实现了多信息融合的人体目标跟踪算法。实验结果表明,在连续自适应均值漂移算法中引入深度信息和预测信息,消除了复杂环境中相似色度和光线对目标的影响,提高了算法的鲁棒性。

Moving object tracking algorithm fused with depth information,color information and forecast information is presented in this paper,which enables intelligent video surveillance system to track motion human object robustly. Firstly,the target model of the algorithm is achieved using the color information and depth information fusion,which is integrated into the Camshift tracking algorithm when the targets are tracked. Then the search window is predicted by Kalman filter. And the target centroid is calculated in the corresponding depth image. Finally,object tracking algorithm fused with Multi-information is realized using Kinect sensor. The tests show that the influence of similar color and light on the target is eliminated in the complex environment by introducing the depth information and prediction information in a continuous adaptive mean shift algorithm,thus improving the robustness of the algorithm.