應用Mask R-CNN於路面速限標記之偵測

APPLY MASK R-CNN TO THE DETECTION OF ROAD SPEED LIMIT MARKINGS

謝宜均(Yi-Chun Hsieh) ； 呂學展(Eric Hsueh-Chan Lu) ； 邱靜梅(Jing-Mei Ciou)

《中國土木水利工程學刊》 34卷3期 (2022/05) Pp. 221-228

https://doi.org/10.6652/JoCICHE.202205_34(3).0005

摘要

近年來，自動駕駛產業蓬勃發展，越來越多人投入相關的研究，自駕車需對道路上的物件進行準確的偵測和分類，以確保行車的穩定性。其中電腦視覺、影像辨識以及基於深度學習的物件偵測是自動駕駛的關鍵技術。先前研究對於道路上的物件偵測多著重於路牌辨識、交通號誌等物件，較少有路面標記的研究。本團隊先前已將道路模型分類為11類來進行道路物件偵測，但實驗結果的類別信心值下降且有辨識錯誤的情形發生，本團隊推斷是因為訓練類別太多導致模型學習太多特徵而致。在本論文中，我們使用著名的實例分割（Instance Segmentation）模型，Mask R-CNN來進行模型的訓練，並將路面速限類別獨立訓練成一模型，訓練完的模型於辨識物件的準確度及類別信心值皆有顯著提升。速限物件偵測模型在未來也可配合點雲作為高精地圖的建置，以健全台灣自駕車產業發展。

關鍵字

深度學習 ； 實例分割 ； 路面速限標誌

並列摘要

Self-driving cars industry has been booming in recent years. The accurate detection and classification of road objects are required for self-driving cars to ensure the driving stability. Computer vision, image recognition and deep-learning-based object detection are the main technologies for self-driving cars. Previous studies have focused on the detection of traffic signs, traffic signals and so on, but road markings have received less attention. Our team has classified the road model into 11 classes, but the confidence scores of the experiment results have decreased. We inferred that there were too many training classes, causing the model to learn too many features. In this study, we train the model using the well-known Instance segmentation Mask R-CNN and independently train the class of road speed limit markings into a model. The trained model has improved significantly on the accuracy and confidence scores of the object.

並列關鍵字

deep learning ； instance segmentation ； road speed limit markings

參考文獻

Krizhevsky, A., Sutskever, I., and Hinton, G. E., “Imagenet classification with deep convolutional neural networks,” Advances in Neural Information Processing Systems, Vol. 25, pp. 1097-1105 (2012).

He, K., Zhang, X., Ren, S., and Sun, J., “Deep residual learning for image recognition,” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778 (2016).

Krizhevsky, A., Sutskever, I., and Hinton, G. E., “Imagenet classification with deep convolutional neural networks,” Advances in Neural Information Processing Systems, Vol. 25, pp. 1097-1105 (2012).

Girshick, R., “Fast r-cnn,” Proceedings of the IEEE International Conference on Computer Vision, pp. 1440-1448 (2015).

Ren, S., He, K., Girshick, R., and Sun, J., Faster r-cnn: Towards real-time object detection with region proposal networks, Advances in Neural Information Processing Systems, Vol. 28, pp. 91-99 (2015).