基于双向长短期记忆网络及注意力机制的室内行人模式识别

浏览次数：16

分享到:

摘要：

在室内空间进行准确的行人模式识别/场景感知,特别是与位置关联的识别/感知,对于行人的聚集或追踪具有重要意义.针对传统机器学习方法特征提取困难、分类精度低,非正常性行为造成较大识别误差等问题,提出一种基于注意力机制和双向长短记忆(bidirectional long short-term memory,Bi-LSTM)网络的室内实时行人模式识别的模型.建立Bi-LSTM网络提取滑动窗口内行人模式时序特征,评估模型网络结构的性能与时效性,优化所提网络的Bi-LSTM层数和隐藏层节点数,并确定最优的网络结构;为了削减噪声数据对模型的影响,提高网络筛选信息特征的能力,引入注意力机制对所提取的时序特征进行权重参数优化.实验结果表明,相比传统机器学习算法,优化参数后的Bi-LSTM网络,行人模式识别准确度平均提高6.37%,进一步引入注意力机制后,识别准确度平均提高9.21%,最终准确度可达99.32%.所提模型可以有效对行人模式/场景感知进行分类,为室内精准定位追踪提供方法支持.

In indoor space,accurate pedestrian pattern recognition/scene perception,especially the recognition/perception associated with location,is of great significance for pedestrian gathering or tracking.Aiming at the problems of traditional machine learning methods such as difficulty in feature extraction,low classification accuracy,and large recognition errors caused by abnormal behaviors,this paper proposes an indoor real-time pedestrian pattern recognition model based on attention mechanism and bidirectional long short-term memory(Bi-LSTM)network.The Bi-LSTM network was established to extract the temporal characteristics of the pedestrian mode in the sliding window,evaluate the performance and timeliness of the model network structure,optimize the number of Bi-LSTM layers and the number of hidden layer nodes,and determine the optimal network structure.In order to reduce the influence of noise data on the model and improve the ability of the network over screen information features,the attention mechanism is introduced to optimize the weight parameters of the extracted temporal features.The experimental results show that compared with the traditional machine learning algorithm,the accuracy of pedestrian pattern recognition in the optimized Bi-LSTM network is improved by 6.37%on average.After further introducing the attention mechanism,the accuracy of pedestrian pattern recognition is improved by 9.21%on average,and the final accuracy can reach 99.32%.The proposed model can effectively classify the pedestrian mode/scene perception,and provide method support for accurate indoor positioning and tracking.

作者：

梁玉杰崔博

Liang Yujie;Cui Bo(Artificial Intelligence,North China University of Science and Technology,Tangshan 063210,China;Hebei Key Laboratory of Industrial Intelligent Perception North China University of Science and Technology,Tangshan 063210,China)

机构地区：

华北理工大学人工智能学院华北理工大学河北省工业智能感知重点实验室

引用本文：

《betway官方app 学报（自然科学版）》 CAS 北大核心 2024年第3期88-97,共10页

Journal of Henan Normal University(Natural Science Edition)

基金：

2021年度教育部产学合作协同育人项目(202101138019) 2021年度教育部高等学校电子信息类专业教学指导委员会项目(2021-JG-04).

关键词：

行人模式识别滑动窗口时序特征 Bi-LSTM 注意力机制

pedestrian pattern recognition sliding window temporal feature Bi-LSTM mechanism of attention

分类号：

TP391 [自动化与计算机技术—计算机应用技术]