摘要： 针对深度图像中的人体目标跟踪问题，提出一种基于深度图像的改进Camshift算法。利用人体目标的深度信息计算概率分布，结合人体形态学特征，对深度的概率分布赋予不同的权重，通过Camshift算法进行迭代，从而寻找目标，使用卡尔曼滤波器在三维空间中对运动人体目标的位置实现预测和更新。采集1 200帧图像进行测试，结果表明，该算法能实时准确地跟踪深度图像中的运动人体目标，有效克服遮挡等干扰，单人和双人跟踪准确率均在95%以上，高于传统Camshift算法。
关键词:
深度图像,
人体跟踪,
Camshift算法,
卡尔曼滤波器,
核函数,
形态学
Abstract: This paper proposes an improving Camshift algorithm based on depth data in order to realize real-time human body objects tracking in depth image. This algorithm computes depth probability distribution function of human body objects, combining the morphological characteristics of people. Different weight factors are given to the different part of human on depth probability distribution function. It finds human body objects in a frame after several times of iterations, uses the modified Camshift algorithm. Kalman filter is also applied in this work to predict the position of people in 3D space. Doing experiments on 1 200 frames of depth image, results present that this algorithm are effective to track moving human body on depth image even though the objects are partly covered or the shapes are regular changed. For the common one or two people situation, the tracking accuracy rate is over 95%, which is better than traditional Camshift algorithm.
Key words:
depth image,
human body tracking,
Camshift algorithm,
Kalman filter,
kernel function,
morphology
中图分类号: