基于混合跟踪模型的室内步行人体3D运动估计

针对步行人体3D运动估计过程中的自遮挡问题, 提出了基于混合跟踪模型的粒子滤波算法. 首先, 利用自遮挡状态检测模型, 将步行人体运动划分为四种自遮挡状态; 其次, 根据混合跟踪模型, 针对不同的自遮挡状态, 算法采用不同的跟踪模型; 最后, 为了估计遮挡状态下的人体运动, 算法提出了基于M-估计的在线训练方法 以训练肢体运动相关系数. 经过实验分析, 算法对处 于自遮挡状态下的人体3D运动估计有着良好的效果, 人体3D运动的估计精度得到了提高.

Estimating the Pedestrian 3D Motion Indoor via Hybrid Tracking Model

Focusing on self-occlusion of pedestrian 3D motion estimation, the paper proposed a hybrid tracking model particle filter algorithm. First, using the self-occlusion state detecting model, the pedestrian motion can be detected and divided into four self-occlusion states. Second, via hybrid tracking model, different tracking patterns are proposed to track pedestrian motion on different self-occlusion states. And finally, for estimating the pedestrian motion on the self-occlusion state, we proposed the on-line training algorithm based on M-estimate to train the limbs motion correlation coefficient. The result of experiment showed that our algorithm acquires good results of estimating pedestrian 3D motion on the occlusion state and advances the accuracy of estimating pedestrian 3D motion.