透過您的圖書館登入
IP:18.117.158.174
  • 學位論文

深度學習應用於交通標誌及燈號偵測

Deep Learning Applied To Traffic Signs And Signal Detection

指導教授 : 丁肇隆

摘要


人工智慧的發展近年來突飛猛進,幾乎在各個領域都有其發揮的空間而且效果顯著,最明顯的例子之一就是自駕車的發展。車身配備以雷達、電腦視覺、GPS等技術分析車子周遭環境,判斷行進路線是否偏移、與前車的距離、交通標誌變化、及是否有障礙物或其他突發狀況等等。但是現階段離L5級的完全自動駕駛還有一大段距離,就算是L3級自動駕駛,目前也只適合應用在有條件的獨立軌道,如捷運、高鐵等大眾運輸工具,更遑論完全自駕車的普及。現在各車廠推出的新車最多也只有到L2級的程度,行車仍然得依靠駕駛判斷當時的情況做適當的反應。為了減少因駕駛本身疏忽違反交通標誌而釀災,本研究針對紅綠燈和部分交通標誌,以電腦視覺、影像辨識、卷積神經網路(Convolution Neural Network)及Fast R-CNN(Fast Region-based Convolution Neural Network)等技術為基礎,設計一套駕駛輔助系統,以實現交通標誌及燈號的提醒與紅綠燈號轉變時的警告,並且提供簡單有效的過濾方法,降低因偵測錯誤輸出錯誤警告的風險,在道路實測上取得良好成果。

並列摘要


The development of artificial intelligence has grown by leaps and bounds in recent years. It has its own space and effect in almost every field. One of the most obvious examples is the development of self-driving cars. The body is equipped with radar, computer vision, GPS and other technologies to analyze the surrounding environment of the car, to determine whether the route is offset, the distance from the preceding car, the change of traffic signs, and whether there are obstacles or other emergencies. However, at this stage, there is still a long distance from the fully automatic driving of the L5 class. Even the L3 automatic driving is currently only suitable for use in conditional independent tracks, such as MRT, high-speed rail and other mass transit vehicles, let alone fully self-driving. Popularity. At present, the new cars introduced by various car manufacturers are only up to the L2 level. The driving still has to rely on the driving to judge the situation at that time to respond appropriately. In order to reduce the risk of driving violations of traffic signs by the driver itself, this study is aimed at traffic lights and some traffic signs, including computer vision, image recognition, Convolution Neural Network and Fast R-CNN (Fast Region-based Convolution). Based on technologies such as Neural Network, a driver assistance system is designed to realize the warning of traffic signs and lights and the warning when the traffic lights change. And provide a simple and effective filtering method to reduce the risk of outputting false alarms due to detection errors, and achieve good results in road measurement.

參考文獻


[1] Krizhevsky, A., et al. (2012). ImageNet classification with deep convolutional neural networks. Proceedings of the 25th International Conference on Neural Information Processing Systems - Volume 1. Lake Tahoe, Nevada, Curran Associates Inc.: 1097-1105.
[2] Szegedy, C., et al. (2015). Going deeper with convolutions. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
[3] Hinton, G. E., et al. (2012). "Improving neural networks by preventing co-adaptation of feature detectors." CoRR abs/1207.0580.
[4] Girshick, R. B., et al. (2014). "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation." 2014 IEEE Conference on Computer Vision and Pattern Recognition: 580-587.
[5] Girshick, R. (2015). Fast R-CNN. Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV), IEEE Computer Society: 1440-1448.

延伸閱讀