以深度學習YOLOv3進行雪山隧道之車輛偵測和分類

Vehicle Detection and Classification Hsuehshan Tunnel with Deep Learning VOLOv3

蔡俊明(Chun-Ming Tsai) ； 魏家瑜(Jia-Yu Wei)

《國教新知》 65卷4期 (2018/12) Pp. 43-61

https://doi.org/10.6701/TEEJ.201812_65(4).0004

摘要

雪山隧道每逢假日一定塞車，甚至回堵到宜蘭礁溪市區，造成市區亦跟著塞車，讓要從宜蘭回台北的用路人，塞車變成夢靨。為了幫雪山隧道用路駕駛，以及解決市區塞車，雪山隧道急需一套智慧型車輛偵測系統，來偵測雪山隧道的車輛和車流量，及時提供用路駕駛知道，以利提早改道。本研究應用深度學習的YOLOv3來訓練、偵測和辨識雪山隧道的車輛，同時，為了解決原始YOLOv3所提供的訓練權重模型，無法偵測車身不完整的車輛，本研究標註時採4類車型標註，經過、偵測和辨識，雖然目前車輛偵測結果僅80.51%，但是在車輛分類正確率為99.36%。另外，本研究也發現一些問題，未來將繼續努力，來解決這些問題，進而完成雪山隧道智慧型車輛偵測系統。

關鍵字

深度學習YOLOv3 ； 機器學習 ； 車輛偵測和辨識

並列摘要

Hsuehshan Tunnel must be jammed every holiday, and even blocked into the city of Yilan Jiaoxi, causing the city to follow the traffic jam, so that the drivers who want to return to Taipei from Yilan, the traffic jam becomes a nightmare. In order to help Hsuehshan Tunnel to drive and solve the traffic jam in the urban area, Hsuehshan Tunnel urgently needs an intelligence vehicle detection system to detect the vehicle and traffic flow in Hsuehshan Tunnel, and provide real time traffic information for drivers. This study uses deep learning YOLOv3 to train, detect and recognize vehicles in Hsuehshan Tunnel. At the same time, in order to solve the pre-training weight model provided by the original YOLOv3, it is impossible to detect vehicles with incomplete vehicle bodies. After training, detecting and recognizing, although the current vehicle detection result is only 80.51%, the correct classification rate in the vehicle is 99.36%.In addition, this study also found some problems, and will continue to work hard to solve these problems in the future, and then complete the intelligence vehicle detection system in Hsuehshan Tunnel.

並列關鍵字

deep learning YOLOv3 ； machine learning ； vehicle detection and classification

參考文獻

Cai, Z.W.,Saberian, M.,Vasconcelos, N.(2015).Learning complexity-aware cascades for deep pedestrian detection.Proceedings of the IEEE International Conference on Computer Vision.(Proceedings of the IEEE International Conference on Computer Vision).

Girshick, R.,Donahue, J.,Darrell, T.,Malik, J.(2014).Rich feature hierarchies for accurate object detection and semantic segmentation.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.(Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition).

He, C.H.,Lam. K.M.(2018).Fast Vehicle Detection with Lateral Convolutional Neural Network.IEEE ICASSP 2018.(IEEE ICASSP 2018).

LiWang, Y.L.,Wang, H.,Zheng, Y.B.,Ye, H.,Xue, X.Y.(2017).Evolving boxes for fast vehicle detection.Proceedings of the IEEE International Conference on Multimedia and Expo.(Proceedings of the IEEE International Conference on Multimedia and Expo).

Redmon, J.,Farhadi, A.(2018).Yolov3: An incremental improvement.,未出版.