基于时间自适应算法的双路径激励列车驾驶员行为识别方法  

Dual-path motivated train driver behavior recognition method based on time adaptive algorithm

作　　者：李孟珍 孟子诤 王伟[3] 崔文利 刘明亮[1] LI Mengzhen;MENG Zizheng;WANG Wei;CUI Wenli;LIU Mingliang(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China;General Manager Office,National Railway Printing Company Limited,Beijing 101102,China;Meteorological Data Center,Heilongjiang Meteorological Bureau,Harbin 150030,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080 [2]国铁印务有限公司总经办,北京101102 [3]黑龙江省气象局气象数据中心,哈尔滨150030

出　　处：《黑龙江大学自然科学学报》2024年第5期579-589,共11页Journal of Natural Science of Heilongjiang University

基　　金：黑龙江省自然科学基金资助项目(LH2020F046)。

摘　　要：随着人们出行需求的增加,列车逐渐成为日常出行不可或缺的交通工具,列车驾驶员在途行为对保障安全出行具有至关重要的意义。为降低驾驶员值乘过程中可能出现的操作风险,针对列车驾驶员在途异常行为进行检测,提出了一种基于时间自适应算法的双路径激励驾驶员行为识别方法。该方法通过在时空和通道等维度变换,实现视频帧间的时间信息交流。采用时空激励模块和通道激励模块分别用于激励视频间时空和通道特征,时间自适应模块用于处理输入不同视频流。采用自制数据集、公开行为识别数据集UCF101和Something-Something-v1评估提出方法的性能,在自制数据集达到95.99%的准确率,在UCF101数据集达到96.10%的准确率,在Something-Something-v1数据集上也取得了46.30%的准确率,验证了方法的有效性和泛化性。With the continuous increase of people’s travel demand,the train has gradually become an indispensable means of transportation for people’s daily travel,and the behavior of train drivers on the way is of vital significance to ensure the safety of travel.In order to reduce the possible operational risks in the process of driver duty,the time adaptive dual-path motivated train driver behavior recognition method is proposed to detect the abnormal behavior of train drivers in transit.This method realizes the temporal information exchange between video frames by transforming them in spatio-temporal,channel and other dimensions.The spatio-temporal excitation module and the channel excitation module are used to excite the spatiotemporal and channel features between videos,respectively.The time adaptive module is used to process different input video streams.The self-made dataset,the public behavior recognition dataset UCF101 and Something-Something-v1 were used to evaluate the performance of the proposed method,and the accuracy of the proposed method reached 95.99% in the self-made dataset,96.10% in the UCF101 dataset,and 46.30% in the Something-Something-v1 dataset,which verified the effectiveness and generalization of the method.

关 键 词：驾驶员行为识别 时空激励 通道激励 时间自适应算法 

分 类 号：U285[交通运输工程—交通信息工程及控制]