基於深度學習語義分割之城市道路汽車轉向操控

A Deep Learning Based Semantic Segmentation Approach for Car Steering on Urban Roads

塗國星(Kuo-Hsin Tu)

指導教授 ： 傅立成

共同指導教授 ： 蕭培墉

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201703573

摘要

在視覺式自動駕駛系統中，感知與控制是兩個重要且待解決的議題。此外，由於深度卷積神經網路在解決感知與控制問題上有非常好能力，使得深度卷積神經網路成為視覺式自動駕駛系統的解決方案之一。在本論文中，我們證明語義分割可以用來提升視覺式自動駕駛系統的效能。論文中提出了一個使用語義感知並基於端對端深度卷積神經網路的方法來解決自動駕駛中的視覺式控制問題。所提出的方法具有兩個階段並透過影像輸入來預測汽車轉向操控。在第一個階段中，使用一個深度卷積神經網路從輸入影像產生語義分割的結果，在第二個階段中則使用另一個深度卷積神經網路從語義分割資訊來預測出汽車轉向操控。在實驗中，我們使用一個公開的汽車駕駛資料集來評估所提出的方法，實驗結果顯示該方法能達到比一般端對端的深度卷積神經網路方法更好的結果。

關鍵字

深度學習 ； 卷積神經網路 ； 語義分割 ； 自動駕駛 ； 車輛轉向

並列摘要

In vision based autonomous driving systems, perception and control tasks are two critical problems to be solved. The effectiveness of deep convolutional neural networks (CNNs) in solving visual perception and control tasks has made CNNs a desirable solution for autonomous driving. In this thesis, we show that semantic segmentation can be applied to enhance the performance of a vision based autonomous driving system. We propose an end-to-end CNN architecture with semantic perception to solve the vision based control problem in autonomous driving. The proposed approach is a two-stage CNN architecture that takes a monocular image and outputs a steering angle. In the first stage, a CNN module is used to generate semantic segmentation from the input image. In the second stage, another CNN module is used to take advantage of the semantic perception to predict steering angles. In the experiment, a publicly available dataset of human driving data is used to evaluate the proposed method. Experimental results demonstrate that the proposed method enhance the results of the typical end-to-end CNN approach.

並列關鍵字

Deep learning ； Convolutional neural networks ； Semantic segmentation ； Autonomous driving ； Vehicle steering

參考文獻

[1] Global Status Eport on Road Safety 2015. Available: http://www.who.int/violence_injury_prevention/road_safety_status/2015/en/

[3] V. Sze, Y.-H. Chen, T.-J. Yang, and J. Emer, "Efficient Processing of Deep Neural Networks: A Tutorial and Survey," arXiv preprint arXiv:1703.09039, 2017.

[4] S. Ren, K. He, R. Girshick, and J. Sun, "Faster R-Cnn: Towards Real-Time Object Detection with Region Proposal Networks," in Advances in Neural Information Processing Systems, pp. 91-99, 2015.

[5] J. Long, E. Shelhamer, and T. Darrell, "Fully Convolutional Networks for Semantic Segmentation," in IEEE Conference on Computer Vision and Pattern Recognition, pp. 3431-3440, 2015.

[6] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet Classification with Deep Convolutional Neural Networks," in Advances in Neural Information Processing Systems, pp. 1097-1105, 2012.