一种基于筛选机制的快速概率占据图目标定位算法

A Sifting Mechanism Based Object Localization Algorithm for Fast Probabilistic Occupancy Map

提出一种基于筛选机制的快速概率占据图目标定位算法(SPOM),在多视角监控环境下,该方法能够快速准确地计算出进入场景中运动物体的位置.具体而言,首先设计了一种高效的筛选机制,可以根据运动检测的结果,粗略估计出运动目标在3维空间中的位置;然后建立合适的似然模型,利用贝叶斯方法计算出目标出现在备选区域内各个位置上的概率,从而找到目标物体;最后,通过阈值化概率图的方法得到目标的位置信息,并采用粒子滤波器对定位结果进行校正,以进一步提高定位的准确度.相较于通常的概率占据图算法,该算法通过引入筛选机制来筛除目标不可能出现的位置,可大幅减小概率占据图的计算量,提高了运行速度,并且能够更准确地计算出目标物体的位置.基于自行搭建的实验平台,对这种基于筛选机制的定位算法和通常的概率占据图算法进行了对比实验,实验结果验证了本文算法能够更加快速准确地估计出动态目标的位置.

A sifting mechanism based object localization algorithm for fast probabilistic occupancy map（sifted probabilistic occupancy map, SPOM） is proposed to calculate the positions of moving objects fast and accurately in typical multi-view surveillance scenarios. Specifically, an efficient sifting mechanism is designed firstly to roughly estimate the 3D positions of the moving objects according to the output of motion detection. Secondly, a proper likelihood model is set up by Bayesian method to calculate the occupancy probability of the objects for each position within the sifted region. Finally, object positions are obtained according to a pre-set threshold of probabilistic occupancy map, and particle filter is utilized to adjust the results to improve the localization accuracy. Compared with the conventional probabilistic occupancy map（POM）, the proposed method can decrease the computational overload dramatically by discarding the impossible object positions through the sifting process, therefore the running speed is improved, and it simultaneously provides more accurate estimations for object positions. Based on the self-built experimental platform, comparative experiments of SPOM and POM are conducted,and the obtained results demonstrate that the proposed method can locate the moving objects more quickly and accurately.