| 摘 要: 行为预测会面临行人部位遮挡、背景干扰、摄像机视角不同、行人姿态不同、外观差异过大及行人动作信息提取难度过大等问题。文章提出一种新的基于网格划分骨骼的行为预测方法,首先使用自下而上的方法提取行人的骨骼信息,通过学习人体关节点的距离度量特征和角度度量特征提取行人的行为特征。然后对关节点分别对比前后帧的行为特征,判断下一帧单个关节点运动类型发生的概率,通过对下一帧关节点运动类型的加权判断下一帧行人的动作。所提方法预测人体右脚关节点向上运动的概率为92.3%。 |
| 关键词: 行为预测;自下而上;关节点;距离度量 |
|
中图分类号: TP181
文献标识码: A
|
|
| Action Prediction Method Based on Grid Partition Skeleton |
|
SHEN Jianglin, WEI Dan, WANG Ziyang
|
(School of Mechanical and Automobile Engineering, Shanghai University of Engineering Science, Shanghai 201620, China)
goatlmljrf@163.com; weidan@sues.edu.cn; wangziyangwilliam@163.com
|
| Abstract: Action prediction faces problems such as pedestrian parts occlusion, background interference, different camera angles, different pedestrian posture, large appearance difference, and difficulty in extracting pedestrian motion information. This paper proposes a new action prediction method based on grid partitioning skeleton. Firstly, the bottom-up method is used to extract the bone information of pedestrians, and then the action characteristics of pedestrians are extracted by learning the distance measurement features and angle measurement features of human joints. By comparing human joints action characteristics of two consecutive frames, the probability of the movement type of a single joint in the next frame is determined, and the movement of pedestrians in the next frame is determined by weighting the movement type of the next frame. The probability of predicting the upward movement of the right foot joint by the proposed method is 92.3% . |
| Keywords: action prediction; bottom-up; key point; distance measure |
