针对林区环境中通过监控视频统计车流量的传统方法提取车辆特征困难、无法分类统计等问题,提出了一种基于YOLOv5结合DeepSORT的车流量分类统计方法。该方法使用目标检测算法YOLOv5作为检测器对车辆进行分类检测,为了提升实际场景中的车辆检测效果,在算法中融入CBAM注意力机制增强检测器对车辆的特征提取能力,同时将NMS改进为DIoU-NMS,解决了因车辆相互遮挡导致的漏检问题。使用目标跟踪算法DeepSORT对检测到的车辆进行跟踪,为了减少车辆身份切换现象,将重识别网络在车辆重识别数据集上重新训练。最后通过在视频中设置虚拟线的方法对跟踪到的车辆进行统计。将该方法在实际场景中进行效果验证,实验结果表明,总体车流量统计准确率较改进前提升10.1%,汽车、货车、客车的车流量统计准确率分别为91.8%,94.6%,93.8%。
Abstract
In view of the problems such as difficulty in extracting vehicle features and inability to classify statistics by the traditional method of statistical traffic flow through surveillance video in forest environment, a method of traffic flow classification statistics based on YOLOv5 combined with DeepSORT was proposed. The method used the objective detection algorithm YOLOv5 as a detector to classify and detect vehicles. In order to improve the vehicle detection effect in the actual scene, the CBAM attention mechanism was incorporated into the algorithm to enhance the feature extraction ability of the detector for vehicles. In addition, the NMS was improved to DIoU-NMS so as to solve the problem of missed detection caused by mutual vehicle occlusion. The objective tracking algorithm DeepSORT was used to track the detected vehicles, and the reidentification network was retrained on the vehicle re-identification dataset in order to reduce the vehicle identity switching phenomenon. Finally, the tracked vehicles were counted by setting virtual lines in the video. The results of the method were verified in the actual scenario. As shown by the experimental results, the overall traffic flow statistics accuracy was improved by 10.1% compared with that before the improvement. Besides, the traffic flow statistics accuracy of cars, trucks, and buses reached 91.8%, 94.6% and 93.8% respectively.
关键词
计量学 /
车流量统计 /
目标检测 /
目标跟踪;视频监控;YOLOv5算法;DeepSort算法;图像处理
Key words
metrology /
traffic statistics /
object detection /
object tracking /
video surveillance;YOLOv5 algorithm /
DeepSort algorithm /
image processing
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
［1］田为广,徐海黎,殷戎飞,等.基于地磁传感器的车流量智能检测系统设计［J］.传感技术学报,2021,34(1):137-142.
Tian W G, Xv H L, Yin R F, et al. Design of Intelligent Traffic Flow Detection System Based on Geomagnetic Sensor［J］. Chinese Journal of Sensors and Actuators, 2021,34(1):137-142.
［2］丁苏楠, 张秋菊. 基于改进SIFT算法的图像匹配方法［J］. 传感器与微系统, 2020, 39(10):45-47.
Ding S N,Zhang Q J. Image registration method based on improved SIFT algorithm ［J］. Transducer and Microsystem Technologies, 2020,39(10):45-47.
［3］Kim J, Baek J, Kim E. A Novel On-Road Vehicle Detection Method Using πHOG［J］. IEEE transactions on intelligent transportation systems, 2015, 16(6):3414-3429.
［4］Lekha S , Suchetha M . A Novel 1-D Convolution Neural Network With SVM Architecture for Real-Time Detection Applications［J］. IEEE Sensors Journal, 2017, 18(99):724-731.
［5］Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation［C］//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014:580-587.
［6］He K, Zhang X, Ren S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition［J］. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9):1904-1916.
［7］Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector［C］//European conference on computer vision. Springer, Cham, 2016:21-37.
［8］Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection［C］//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.
［9］Bewley A, Ge Z, Ott L, et al. Simple online and realtime tracking［C］//2016 IEEE international conference on image processing (ICIP). IEEE, 2016: 3464-3468.
［10］Wojke N, Bewley A, Paulus D. Simple online and realtime tracking with a deep association metric［C］//2017 IEEE international conference on image processing (ICIP). IEEE, 2017: 3645-3649.
［11］张瑶, 卢焕章, 张路平, 等. 基于深度学习的视觉多目标跟踪算法综述［J］. 计算机工程与应用, 2021, 57(13):55-66.
Zhang Y, Lu H Z, Zhang L P, et al. Overview of Visual Multi-object Track Algorithms with Deep Learning［J］. Computer Engineering and Applications, 2021,57(13):55-66.
［12］Kim S J, Nam J Y, Ko B C. Online tracker optimization for multi-pedestrian tracking using a moving vehicle camera［J］. IEEE Access, 2018, 6:48675-48687.
［13］陈佳倩, 金晅宏, 王文远, 等. 基于YOLOv3和DeepSort的车流量检测［J］. 计量学报, 2021, 42(6):718-723.
Chen J Q, Jin X H, Wang W Y, et al. Vehicle Flow Detection Based on YOLOv3 and DeepSort［J］. Acta Metrologica Sinica, 2021, 42(6):718-723.
［14］Guo M H, Xu T X, Liu J J, et al. Attention mechanisms in computer vision: A survey［J］. Computational Visual Media, 2022:1-38.
［15］Woo S, Park J, Lee J Y, et al. Cbam: Convolutional block attention module［C］//Proceedings of the European conference on computer vision (ECCV). 2018: 3-19.
［16］Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: Faster and better learning for bounding box regression［C］//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(07): 12993-13000.
［17］Wojke N, Bewley A. Deep cosine metric learning for person reidentification［C］//2018 IEEE winter conference on applications of computer vision (WACV). IEEE, Lake Tahoe, NV, USA, 2018:748-756.
［18］程淑红, 谢文锐, 张典范, 等. 基于多算法融合的跌倒行为识别［J］. 计量学报, 2022, 43(1): 107-113.
Cheng S H, Xie W R, Zhang D F, et al. Fall Action Recognition Based on Computer Vision［J］. Acta Metrologica Sinica, 2022, 43(1):107-113.
基金
国家自然科学基金(51968065);云南省教育厅科学研究基金(111722038)
PDF(891 KB)
