| 38 | 0 | 14 |
| 下载次数 | 被引频次 | 阅读次数 |
为解决地铁车站隔栏递物异常行为检测中场景复杂度高、识别难度大和误检率高等问题，本文提出一种基于激光雷达和相机传感器数据融合的检测方法。采用体素差分算法处理激光雷达点云数据，划分检测区域体素单元并建立递物触发机制。采用目标锁定算法融合激光雷达与相机采集的数据，补充人体关键点的深度信息，锁定隔栏递物行为目标物。对时空图卷积网络(spatial-temporal graph convolutional network, STGCN)进行轻量化改造，降低模型复杂度，减少计算时间；引入时间趋势注意力(temporal trend attention, TTA)模型，增强隔栏递物行为姿态时空变化特征的提取能力，形成TTA-STGCN模型，计算隔栏递物行为发生的置信度。通过实验室模拟和地铁车站现场采集隔栏递物数据并制定检测效果评价指标，进行STGCN模型、STGCN-MIN模型、TTA-STGCN模型的训练、验证和测试，在训练阶段，TTA-STGCN模型的准确率比前二者均增大3.73%,整体损失比前二者均减小66.00%;在验证阶段，TTA-STGCN模型的准确率比前二者分别增大3.89%、0.68%,整体损失比前二者分别减小58.95%、58.48%;在测试阶段，TTA-STGCN模型的准确率比前两者均增大3.15%,整体损失比前二者分别减小42.85%、44.40%。进行现场试验，相比STGCN模型、STGCN-MIN模型，TTA-STGCN模型的准确率分别增大2.99%、3.49%;精确率分别增大2.28%、1.31%;召回率分别增大4.30%、6.45%;F1分数分别增大0.033 5、0.040 4,表明TTA-STGCN模型显著提高地铁车站特定场景下隔栏递物行为的检测精度。
Abstract：To address the problems of high scene complexity, difficult recognition, and high false detection rate in detecting abnormal behavior of object passing through subway station barriers, a detection method is proposed that data fusion based on light detection and ranging(LiDAR) and camera sensors. A voxel difference algorithm is used to process LiDAR point cloud data, divide the detection area into voxel units, and establish a trigger mechanism for object passing. An object locking algorithm is employed to fuse data collected by LiDAR and cameras, supplementing depth information of human key points and locking onto target of object passing through barriers. The spatial-temporal graph convolutional network(STGCN) is lightweight-modified to reduce model complexity and computation time. A temporal trend attention(TTA) model is introduced to enhance the extraction of spatial-temporal feature changes in postures of object passing through barriers, forming the TTA-STGCN model to calculate the confidence of behavior occurrence of object passing through barriers. Collecting data of object passing through barriers through laboratory simulation and on-site in subway stations. Detection performance evaluation metrics are established. Training, validation, and testing of STGCN, STGCN-MIN, and TTA-STGCN models are conducted. In the training phase, the accuracy of the TTA-STGCN model improved by 3.73% compared to the first two, the overall loss decreased by 66.00%. In the validation phase, the accuracy of the TTA-STGCN model improved by 3.89% and 0.68% compared to the first two respectively, the overall loss decreasing by 58.95% and 58.48% respectively. In the testing phase, the accuracy of the TTA-STGCN model improved by 3.15% compared to the first two, the overall loss decreasing by 42.85% and 44.40% respectively. Field experiments show that the TTA-STGCN model′s accuracy improved by 2.99% and 3.49% compared to STGCN-MIN and STGCN models respectively, precision improved by 2.28% and 1.31% respectively, recall improved by 4.30% and 6.45% respectively, and F1 score improved by 0.033 5 and 0.040 4 respectively, demonstrating that the TTA-STGCN model significantly enhances the detection accuracy of behavior of object passing through barriers in specific subway station scenarios.
[1] 吴洁.城市轨道交通运营安全管理的有效措施[J].人民公交，2024(20):82-84.
[2] 安俊峰，刘吉强，卢萌萌，等.基于改进YOLOv8的地铁站内乘客异常行为感知[J].北京交通大学学报，2024,48(2):76-89.
[3] RAO A S,GUBBI J,MARUSIC S,et al.Crowd event detection on optical flow manifolds[J].IEEE Transactions on Cybernetics,2016,46(7):1524-1537.
[4] MAJI D,NAGORI S,MATHEW M,et al.YOLO-pose:enhancing YOLO for multi person pose estimation using object keypoint similarity loss[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.New Orleans,LA,USA:IEEE,2022:2636-2645.
[5] YAN S J,XIONG Y J,LIN D H.Spatial temporal graph convolutional networks for skeleton-based action recognition[C]// Proceedings of the AAAI Conference on Artificial Intelligence.New Orleans,Louisiana,USA:AAAI,2018,32(1):7444-7452.
[6] BREHAR R D,MURESAN M P,MARI?A T,et al.Pedestrian street-cross action recognition in monocular far infrared sequences[J].IEEE Access,2021,9:74302-74324.
[7] DU Y,HOU Z J,LI X,et al.PointDMIG:a dynamic motion-informed graph neural network for 3D action recognition[J].Multimedia Systems,2024,30(4):192.
[8] SUN Z H,KE Q H,RAHMANI H,et al.Human action recognition from various data modalities:a review[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(3):3200-3225.
[9] SHI Z S,LIANG J,LI Q Q,et al.Multi-modal multi-action video recognition[C]//Proceedings of IEEE/CVF International Conference on Computer Vision.Montreal,QC,Canada:IEEE,2021:13658-13667.
[10] GUO W Z,WANG J W,WANG S P.Deep multimodal representation learning:a survey[J].IEEE Access,2019,7:63373-63394.
[11] ZHANG D,JU X C,ZHANG W,et al.Multi-modal multi-label emotion recognition with heterogeneous hierarchical message passing[C]//Proceedings of the AAAI Conference on Artificial Intelligence.Pennsylvania,USA:AAAI,2021,35(16):14338-14346.
[12] NIE W Z,YAN Y,SONG D,et al.Multi-modal feature fusion based on multi-layers LSTM for video emotion recognition[J].Multimedia Tools and Applications,2021,80(11):16205-16214.
[13] BLASCH E,PHAM T,CHONG C Y,et al.Machine learning/artificial intelligence for sensor data fusion-opportunities and challenges[J].IEEE Aerospace and Electronic Systems Magazine,2021,36(7):80-93.
[14] QIU S,ZHAO H K,JIANG N,et al.Multi-sensor information fusion based on machine learning for real applications in human activity recognition:state-of-the-art and research challenges[J].Information Fusion,2022,80:241-265.
[15] 董文波.基于激光雷达与相机融合的铁路站场内障碍物检测研究[D].成都：西南交通大学，2022.
[16] LIU H B,WU C,WANG H J.Real time object detection using LiDAR and camera fusion for autonomous driving[J].Scientific Reports,2023,13(1):8056.
[17] OSOKIN D.Real-time 2D multi-person pose estimation on CPU:lightweight OpenPose[EB/OL].(2018-11-29)[2024-12-21].https://arxiv.org/abs/1811.12004v1.
[18] CAO Z,SIMON T,WEI S H,et al.Realtime multi-person 2D pose estimation using part affinity fields[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Honolulu,HI,USA:IEEE,2017:1302-1310.
[19] ZHOU Y,TUZEL O.VoxelNet:end-to-end learning for point cloud based 3D object detection[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City,UT,USA:IEEE,2018:4490-4499.
[20] YU B,YIN H T,ZHU Z X.Spatio-temporal graph convolutional networks:a deep learning framework for traffic forecasting[EB/OL].(2018-07-12)[2024-12-21].https://arxiv.org/abs/1709.04875v4.
[21] LI S Y,JIN X Y,XUAN Y,et al.Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting[EB/OL].(2020-01-03)[2024-12-21].https://arxiv.org/abs/1907.00235v3.
[22] ZHOU H Y,LIU Q J,WANG Y H.Learning discriminative representations for skeleton based action recognition[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Vancouver,BC,Canada:IEEE,2023:10608-10617.
[23] 李磊磊.基于多模态关键语义学习的驾驶环境交通事故预测[D].西安：长安大学，2024.
[24] 张浩晨，张竹林，史瑞岩，等.基于YOLO-NPDL的复杂交通场景检测方法[J].山东交通学院学报，2025,33(2):34-37.
[25] 蒋仕新，邹小雪，杨建喜，等.复杂背景下基于改进YOLOv8s的混凝土桥梁裂缝检测方法[J].交通运输工程学报，2024,24(6):135-147.
基本信息:
DOI：
中图分类号:TP212.9;TN958.98;U298
引用信息:
[1]班魁国,高佼,阮久宏等.基于传感器数据融合的地铁车站隔栏递物检测方法[J].山东交通学院学报,2025,33(03):1-11.
基金信息:
山东省自然科学基金项目(ZR2022QF107)