文献类型：期刊文献

中文题名：一种基于改进GAN的行人轨迹预测算法

英文题名：A PEDESTRIAN TRAJECTORY PREDICTION ALGORITHM BASED ON IMPROVED GAN

作者：谢龙洋[1];张军[2];刘元盛[2];王子钰[2]

第一作者：谢龙洋

机构：[1]北京联合大学北京市信息服务工程重点实验室,北京100101;[2]北京联合大学机器人学院,北京100101

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2025

卷号：42

期号：3

起止页码：238-243

中文期刊名：计算机应用与软件

外文期刊名：Computer Applications and Software

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金项目(61871038,61931012)。

语种：中文

中文关键词：生成对抗网络;注意力机制;人体姿态;社交交互

外文关键词：Generative adversarial network;Attention mechanism;Human posture;Social interaction

摘要：为准确预测出行人未来轨迹,提出一种基于改进GAN的行人轨迹预测模型。该模型以观察者相机运动状态向量、行人姿态信息及行人历史轨迹作为输入;在注意力模块中,采用运动注意力机制来衡量观察者相机运动对行人轨迹的影响,采用姿态注意力机制提取人体姿态中的隐藏特征,并采用交互注意力机制来建模行人之间的社交交互;通过生成对抗网络获取预测轨迹。实验验证表明,所提出的模型精度高于目前已有算法。
In order to accurately predict the future trajectory of pedestrians,a pedestrian trajectory prediction model based on improved GAN is proposed.The model took the observer's camera motion state vector,pedestrian pose information and pedestrian's historical trajectory as input.In the attention module,the motion attention mechanism was used to measure the impact of the observer's camera motion on the pedestrian's trajectory,and the posture attention mechanism was used to extract the hidden features in the human pose.The interactive attention mechanism was used to model the social interaction between pedestrian.The predicted trajectories were obtained through the generative adversarial network.Experimental results show that the proposed model has higher accuracy than the existing algorithms.