臺灣博碩士論文加值系統

當今的許多車道線檢測方法雖有顯著的性能，但缺乏即時偵測的能力。即時偵測的能力對於自動駕駛車十分重要，但面對車載硬體的限制，如何達到即時偵測仍是十分艱鉅的挑戰。
在本文中提出了，基於循環特徵位移之關鍵點實例網路，一種新穎且有效的車道偵測方法。提取出車道中的關鍵點並進行實例分割。藉由減少網路的沙漏塊，提高運算速度。並利用知識蒸餾與循環特徵位移器減少了因縮短網路造成的準確度損失。在目前兩個較為廣泛使用的車道線資料庫TuSimple 與 CULane 上進行的大量的實驗。由實驗結果可知，循環特徵位移之關鍵點實例網路在速度與準確度 上具有良好的效果)在CULane 資料庫實現了高F1-measure(68.52)。速度方面，在在CULane 資料庫可達54FPS，基於NVIDIA GeForce RTX 3090。

Although many of today's lane line detection methods have significant performance, they lack the ability to detect in real time. The ability of real-time detection is very important for autonomous vehicles, but in the face of the limitations of on-board hardware, how to achieve real-time detection is still a very difficult challenge.
In this master's thesis a novel and effective lane detection method, PINET-RES: Point Instance Networks Based on Recurrent Feature-Shift is proposed. Extract key points in the lane and perform instance segmentation on the key points. Improve computing speed by reducing hourglass blocks in the network. Then knowledge distillation and recurrent feature shifter are used to reduce the accuracy loss caused by shortening the network. A large number of experiments was conducted on CULane, a currently widely used lane line databases. It can be seen from the experimental results that the Point Instance Networks Based on Recurrent Feature-Shift for Lane Detection has good results in speed and accuracy. It achieves high F1-measure (68.52) in the CULane database. In terms of speed, it can reach 54FPS on the CULane database, based on NVIDIA GeForce RTX 3090.

摘要…………………………………………………………………………i
Abstract………………………………………………………………………ii
誌謝………………………………………………………………………iii
目錄………………………………………………………………………iv
表目錄………………………………………………………………………vi
圖目錄………………………………………………………………………vii
第一章 緒論……………………………………………………………1
第二章 相關研究 ………………………………………………………3
2.1 與車道偵測相關的卷積神經網路…………………………………3
2.2 深度學習技術………………………………………………………5
2.2.1 類神經網路………………………………………………………5
2.2.2 卷積神經網路…………………………………………………5
2.2.3 批次歸一化(Batch Normalization)…………………………………7
2.2.4 知識蒸餾與注意力蒸餾(attention distillation)…………………7
2.3 SCNN [6]…………………………………………………………………9
2.4 循環特徵位移器(RESA)[7]……………………………………………13
第三章 研究方法………………………………………………………18
3.1 PINet[19]………………………………………………………………18
3.1.1 網路架構………………………………………………………18
3.1.2 損失函數………………………………………………………25
3.2 PINET-RES……………………………………………………………27
第四章 實驗結果與分析…………………………………………………30
4.1 資料庫與實驗環境…………………………………………………30
4.1.1 CULane[6]………………………………………………………30
4.1.2 實驗環境 ………………………………………………………31
4.1.3 超參數設定………………………………………………………31
4.1.4 評估指標 ………………………………………………………32
4.2 消融實驗……………………………………………………………34
4.3 與其他方法之比較…………………………………………………35
第五章 結論與未來展望…………………………………………………36
參考文獻 …………………………………………………………………37
Extended Abstract ……………………………………………………………40
Chapter1. Introduction ………………………………………………………41
Chapter2. Related work ………………………………………………………41
Chapter3. Research methods …………………………………………………42
Chapter4. Experimental results and analysis …………………………………43
Chapter5. Conclusion ………………………………………………………44

[1]John Canny, “A Computational Approach to Edge Detection” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 8, no. 6, pp. 679-698, 1986.
[2]Vijay Gaikwad and Shashikant Lokhande, “Lane Departure Identification for Advanced Driver Assistance” (in English), IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 2, pp. 910-918, Apr 2015, doi: 10.1109/Tits.2014.2347400.
[3]Yann Lecun, Leon Bottou, Yoshua Bengio, and Patrick Haffner, “Gradient-based learning applied to document recognition,” in Proceedings of IEEE, 1998, vol. 86, no. 11, pp. 2278-2324.
[4]Oriol Vinyals Geoffrey Hinton, Jeff Dean, “Distilling the knowledge in a neural network” STAT, 2015.
[5]Nikos Komodakis and Sergey Zagoruyko, “Paying more attention to attention: improving the performance of convolutional neural networks via attention transfer” presented at the ICLR, Paris, France, 2017-06-01, 2017. [Online]. Available: https://enpc.hal.science/hal-01832769.
[6]Xingang Pan, Jianping Shi, Ping Luo, Xiaogang Wang, and Xiaoou Tang, “Spatial As Deep: Spatial CNN for Traffic Scene Understanding”, in arXiv:1712.06080, ed, 2018.
[7]Tu Zheng, Hao Fang, Yi Zhang, Wenjian Tang, Zheng Yang, HaiFeng Liu, and and Deng Cai, “Resa: Recurrent feature-shift aggregator for lane detection” in Proceedings of the AAAI Conference on Artificial Intelligence, 2021, vol. 35, no. 4, pp. 3547-3554.
[8]Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification” presented at the 2015 IEEE International Conference on Computer Vision.
[9]Seokju Lee, Junsik Kim, Jae Shin Yoon, Seunghak Shin, Oleksandr Bailo, Namil Kim, Tae-Hee Lee, Hyun Seok Hong, Seung-Hoon Han, and In So Kweon, “VPGNet: Vanishing Point Guided Network for Lane and Road Marking Detection and Recognition” presented at the International Conference on Computer Vision, 2017.
[10]Yuenan Hou, Zheng Ma, Chunxiao Liu, and Chen Change Loy, “Learning Lightweight Lane Detection CNNs by Self Attention Distillation” in Proceedings of IEEE/CVF International Conference on Computer Vision, Seoul, South Korea, 2019.
[11]Jonah Philion, “FastDraw: Addressing the Long Tail of Lane Detection by Adapting a Sequential Prediction Network” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, doi: 10.1109/CVPR.2019.01185.
[12]Zequn Qin, Huanyu Wang, and Xi Li, “Ultra fast structure-aware deep lane detection” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXIV 16, 2020: Springer, pp. 276-291.
[13]Lucas Tabelini, Rodrigo Berriel, Thiago M. Paixão, Claudine Badue, Alberto F. De Souza, and Thiago Oliveira-Santos, “PolyLaneNet: Lane Estimation via Deep Polynomial Regression” presented at the 2020 25th International Conference on Pattern Recognition, 2021.
[14]Li Xiang, Jun Li, Hu Xiaolin, and Yang Jian, “Line-CNN: End-to-End Traffic Line Detection With Line Proposal Unit}” IEEE Transactions on Intelligent Transportation Systems, vol. 21, 2020, doi: 0.1109/TITS.2019.2890870.
[15]Lucas Tabelini, Rodrigo Berriel, Thiago M. Paixão, Claudine Badue, Alberto F. De Souza, and Thiago Oliveira-Santos, “Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection” 2021.
[16]Hala Abualsaud, Sean Liu, David B. Lu, Kenny Situ, Akshay Rangesh, and Mohan M. Trivedi, “LaneAF: Robust Multi-Lane Detection with Affinity Fields” IEEE Robotics and Automation Letters, vol. 6, pp. 7477-7484, 2021, Art no. 4, doi: 10.1109/LRA.2021.3098066.
[17]Zhiwei Dong, Zhu, Xi,Cao, Xiya,Ding, Ran,Li, Wei,Zhou, Caifa,Wang, Yongliang and Qiangbo Liu, “BezierFormer: A Unified Architecture for 2D and 3D Lane Detection” 2024.
[18]Zhengyang Feng, Shaohua Guo, Xin Tan, Ke Xu, Min Wang, and Lizhuang Ma, “Rethinking efficient lane detection via curve modeling” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17041-17049, doi: 10.1109/CVPR52688.2022.01655.
[19]Yeongmin Ko, Younkwan Lee, Shoaib Azam, Farzeen Munir, Moongu Jeon, and Witold Pedrycz, “Key Points Estimation and Point Instance Segmentation Approach for Lane Detection” IEEE Transactions on Intelligent Transportation Systems, vol. 23, pp. 8949-8958, 2022, doi: 10.1109/TITS.2021.3088488.
[20]Christian Szegedy Sergey Ioffe, “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift” in Proceedings of the 32nd International Conference on Machine Learning, Lille, France, 2015.
[21]Krähenbühl. Philipp and Koltun. Vladlen, “Efficient inference in fully connected crfs with gaussian edge potentials” presented at the Neural Information Processing System, 2011.
[22]Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Deep Residual Learning for Image Recognition” in Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas,, Nevada, USA,, 2016, pp. 770-778.
[23]Sean Bell, C Lawrence Zitnick, Kavita Bala, and Ross Girshick, “Inside-outside net: Detecting objects in context with skip pooling and recurrent neural networks” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2874-2883.
[24]Xiaodan Liang, Xiaohui Shen, Jiashi Feng, Liang Lin, and Shuicheng Yan, “Semantic object parsing with graph lstm” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14, 2016: Springer, pp. 125-143.
[25]Fisher Yu, Dequan Wang, Evan Shelhamer, and Trevor Darrell, “Deep layer aggregation” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2403-2412.
[26]Alejandro Newell, Kaiyu Yang, and Jia Deng, “Stacked hourglass networks for human pose estimation” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part VIII 14, 2016: Springer, pp. 483-499.
[27]Wei Yang, Shuang Li, Wanli Ouyang, Hongsheng Li, and Xiaogang Wang, “Learning feature pyramids for human pose estimation” in proceedings of the IEEE international conference on computer vision, 2017, pp. 1281-1290.
[28]Kaiwen Duan, Song Bai, Lingxi Xie, Honggang Qi, Qingming Huang, and Qi Tian, “Centernet: Keypoint triplets for object detection” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 6569-6578.
[29]Weiyue Wang, Ronald Yu, Qiangui Huang, and Ulrich Neumann, “Sgpn: Similarity group proposal network for 3d point cloud instance segmentation” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2569-2578.
[30]Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Delving deep into rectifiers: Surpassing human-level performance on imagenet classification” in Proceedings of the IEEE international conference on computer vision, 2015, pp. 1026-1034.