臺灣博碩士論文加值系統

近年來，自駕車技術的快速發展使得相關研究更加重視深度學習的進步，許多深度學習網路架構相繼被提出，例如：Fast R-CNN、YOLO、SSDNet和RetinaNet等。然而，由於這些架構的參數量與運算量過於龐大，導致難以應用於嵌入式系統或邊緣計算裝置，因此許多研究致力於如何在嵌入式系統上實現深度學習架構，這些精簡化網路架構也被稱為輕量化神經網路，諸如：MobileNet ShuffleNet、GhostNet和PP-LCNet等。現今的車輛盲區偵測主要著重於判斷車輛前方目標，並僅限於單一視角的畫面，較少針對以駕駛者視野為出發點的盲區，例如：左右A柱區域等，這種取向未能有效應對看不見的狀況，成為當前車輛安全技術上的一大問題。市面上的盲區偵測主要有二種方法。第一種是使用感測器，如：雷達與光達等，通過這些感測器準確測定車輛的方位和距離，但其缺點是成本過高，且容易受到天候或環境因素的干擾。第二種為結合了機器學習和深度學習的影像辨識，如：支援向量機（Support Vector Machine, SVM）和卷積神經網路（Convolution Neural Network, CNN），但這種方法需要大量樣本進行訓練，且龐大的參數量與運算量使其難以在嵌入式系統上實現。有鑑於此，本論文提出了一套針對車輛左右方盲區，並可建置於嵌入式系統或邊緣計算裝置的「模型剪枝輕量化神經網路之數據增強演算法應用於雙鏡頭車輛盲區偵測系統（Distraction Evaluation by Multi-Task Pruning Lightweight Neural Network with Data Enhancement Algorithm for Dual Side Blind Spot Detection System, DEBLIND）」。DEBLIND整合了輕量化神經網路、數據增強、模型剪枝和距離演算法等技術，使系統在資源有限的情況下，能夠高效處理來自多個感測器的數據，並通過各種隨機變換來生成多樣化的訓練樣本，提升模型在不同駕駛環境下的穩定性。此外，通過刪除冗餘參數，使網路更加高效，降低延遲，並且能準確計算車輛與周圍物體之間的距離，幫助駕駛者避免碰撞風險。這使得DEBLIND能夠更準確高效地偵測和處理車輛左右方的盲區，提升駕駛安全性。

Many deep learning network architectures are proposed, such as Fast R-CNN、YOLO、SSDNet, and RetinaNet, due to the rapid progress of self-driving technology in the past decades. However, the above deep learning network architectures are difficult to be applied to embedded systems and edge computing devices due to the large amount of parameters and calculations. Hence, many lightweight neural networks used on embedded systems are proposed, such as MobileNet, ShuffleNet, GhostNet and PP-LCNet. The current vehicle blind spot detection technique almost focuses on the target detection in front of the vehicle with only a single perspective without considering the driver's field of vision, such as the left and right A-pillar areas. Hence, the vehicle blind spot detection is still the important issue of the current vehicle safety technology. The vehicle blind spot detection is divided into sensor detection and image recognition. In the sensor detection, the position and distance of vehicle are calculated by sensors, such as radar and lidar. However, the cost of these sensors is too high. In addition, the sensor detection is prone to be interfered by the weather or surrounding environmental. In the image recognition, it is often used with machine learning and deep learning, such as support vector machine (SVM) and convolution neural network (CNN). However, it is required a large number of samples for training. Moreover, it is difficult to be implemented on embedded systems due to the huge amount of parameters and calculations. Therefore, a distraction evaluation by multi-task pruning lightweight neural network with data enhancement algorithm for dual side blind spot detection system, DEBLIND, was proposed by integrating the lightweight neural networks, data enhancement, model pruning and distance algorithm to address the above issues in this paper. In DEBLIND, the distance between detected object and driver could be calculated more accurately with lower latency by the various random transformations samples and deleting redundant parameters to avoid the traffic accidents. It also showed that DEBLIND we proposed could be applied to the dual side vehicle blind spot detection to improve driving safety.

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
1.1 研究動機與目的 1
1.2 研究方向 2
1.3 論文架構 2
第二章 相關文獻探討 3
2.1 物件檢測概況 3
2.2 物件檢測發展 3
2.3 傳統物件檢測概況 4
2.4 深度學習物件檢測概況 5
2.5 輕量化神經網路概況 7
2.6 數據增強概況 8
2.7 模型壓縮概況 11
2.8 車輛盲區檢測概況 15
第三章 輕量化神經網路架構 19
3.1 輕量化神經網路架構 19
3.1.1 Mo-GhostNet輕量化神經網路 21
3.1.2 Sh-GhostNet輕量化神經網路 26
3.1.3 PP-GhostNet輕量化神經網路 30
3.2 物件檢測之損失函數 32
3.3 數據增強方法 33
3.4 模型剪枝方法 41
第四章 雙鏡頭車輛盲區偵測系統 43
4.1 雙鏡頭偵測 44
4.2 車輛盲區偵測系統 45
4.2.1 相機視角轉換演算法 46
4.2.2 圖像坐標轉像素坐標 47
4.2.3 相機坐標轉圖像坐標 48
4.2.4 世界坐標轉相機坐標 49
4.2.5 距離演算法 50
第五章 實驗結果與分析 51
5.1 訓練平台 51
5.2 嵌入式系統平台 52
5.3 鏡頭規格 53
5.4 鏡頭架設位置 54
5.5 輕量化神經網路實驗結果分析 55
5.6 數據增強實驗結果分析 57
5.7 模型剪枝實驗結果分析 58
5.8 嵌入式系統效能分析 60
5.9 距離演算法實驗結果分析 61
第六章 結論及未來展望 63
參考文獻 64
附錄 69

[1] "交通部道安資訊查詢網" https://roadsafety.tw/Dashboard/Custom?type=%E7%B5%B1%E8%A8%88%E5%BF%AB%E8%A6%BD
[2] "全國112年車輛左右轉發生事故件數統計" https://roadsafety.tw/AccCauseOrder?type=%E5%A4%A7%E5%9E%8B%E8%BB%8A%E5%B7%A6%E5%8F
[3] R. Girshick, "Fast R-CNN," in Proc. of IEEE International Conference on Computer Vision, Santiago, Chile, pp. 1440-1448, 07-13 December 2015.
[4] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, pp. 779-788, 27-30 June 2016.
[5] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg, "SSD: Single Shot MultiBox Detector," in Proc. of European Conference on Computer Vision, Amsterdam, Netherlands, pp. 21-37, 11-14 October 2016.
[6] T. -Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, "Focal Loss for Dense Object Detection," in Proc. of IEEE International Conference on Computer Vision, Venice, Italy, pp. 2999-3007, 22-29 October 2017.
[7] C. Tang, Y. Feng, X. Yang, C. Zheng, and Y. Zhou, "The Object Detection Based on Deep Learning," in Proc. of International Conference on Information Science and Control Engineering, Changsha, China, pp. 723-728, 21-23 July 2017.
[8] R. Chauhan, K. K. Ghanshala, and R. C. Joshi, "Convolutional Neural Network (CNN) for Image Detection and Recognition," in Proc. of International Conference on Secure Cyber Computing and Communication, Jalandhar, India, pp. 278-282, 15-17 December 2018.
[9] R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, pp. 580-587, 23-28 June 2014.
[10] S. Ren, K. He, R. Girshick, and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 39, No. 6, pp. 1137-1149, Jun. 2017.
[11] N. Dalal, and B. Triggs, "Histograms of Oriented Gradients for Human Detection," in Proc. of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Diego, CA, USA, pp. 886-893, 20-25 June 2005.
[12] S. Zhang, C. Chi, Y. Yao, Z. Lei, and S. Z. Li, "Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, pp. 9756-9765, 13-19 June 2020.
[13] K. Duan, S. Bai, L. Xie, H. Qi, Q. Huang, and Q. Tian, "CenterNet: Keypoint Triplets for Object Detection," in Proc. of IEEE International Conference on Computer Vision, Seoul, Korea, pp. 6568-6577, 27 October 2019 - 02 November 2019.
[14] K. He, G. Gkioxari, P. Dollár, and R. Girshick, "Mask R-CNN," in Proc. of IEEE International Conference on Computer Vision, Venice, Italy, pp. 2980-2988, 22-29 October 2017.
[15] Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto and Hartwig Adam, "MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications," Computer Vision and Pattern Recognition, pp. 1-9, 17 April 2017. (https://arxiv.org/abs/1704.04861)
[16] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. -C. Chen, "MobileNetV2: Inverted Residuals and Linear Bottlenecks," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 4510-4520, 18-23 June 2018.
[17] A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y. Zhu, R. Pang, V. Vasudevan, Q. V. Le, and H. Adam, "Searching for MobileNetV3," in Proc. of IEEE International Conference on Computer Vision, Seoul, Korea, pp. 1314-1324, 27 October 2019 - 02 November 2019.
[18] X. Zhang, X. Zhou, M. Lin, and J. Sun, "ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 6848-6856, 18-23 June 2018.
[19] N. Ma, X. Zhang, H. Zheng, and J. Sun, "ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design," in Proc. of European Conference on Computer Vision, Munich, Germany, pp. 122-138, 08-14 September 2018.
[20] K. Han, Y. Wang, Q. Tian, J. Guo, C. Xu, and C. Xu, "GhostNet: More Features From Cheap Operations," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, pp. 1577-1586, 13-19 June 2020.
[21] Y. Tang, K. Han, J. Guo, C. Xu, C. Xu, and Y. Wang, "GhostNetV2: Enhance Cheap Operation with Long-Range Attention," in Proc. of Conference on Neural Information Processing Systems, New Orleans, USA, pp.1-14, 28 November 2022 - 09 December 2022.
[22] Cheng Cui, Tingquan Gao, Shengyu Wei, Yuning Du, Ruoyu Guo, Shuilong Dong, Bin Lu, Ying Zhou, Xueying Lv, Qiwen Liu, Xiaoguang Hu, Dianhai Yu and Yanjun Ma, "PP-LCNet: A Lightweight CPU Convolutional Neural Network," Computer Vision and Pattern Recognition, pp. 1-8, 17 September 2021. (https://arxiv.org/abs/2109.15099)
[23] S. Divya, B. Adepu, and P. Kamakshi, "Image Enhancement and Classification of CIFAR-10 Using Convolutional Neural Networks," in Proc. of International Conference on Smart Systems and Inventive Technology, Tirunelveli, India, pp. 1-7, 20-22 January 2022.
[24] L. Deng, "The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]," IEEE Signal Processing Magazine, Vol. 29, No. 6, pp. 141-142, Nov. 2012.
[25] A. Buslaev, A. Parinov, E. Khvedchenya, V. I. Iglovikov, and A. A. Kalinin, "Albumentations: Fast and Flexible Image Augmentations," Information, Vol. 11, No. 2, pp. 1-20, Feb. 2020.
[26] Yu Cheng, Duo Wang, Pan Zhou and Tao Zhang, "A Survey of Model Compression and Acceleration for Deep Neural Networks," Machine Learning, pp. 1-10, 14 June 2020. (https://arxiv.org/abs/1710.09282)
[27] B. Jacob, S. Kligys, B. Chen, M. Zhu, M. Tang, A. Howard, H. Adam, and D. Kalenichenko, "Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 2704-2713, 18-23 June 2018.
[28] S. Zhou, Y. Wang, D. Chen, J. Chen, X. Wang, C. Wang, and J. Bu, "Distilling Holistic Knowledge with Graph Neural Networks," in Proc. of IEEE International Conference on Computer Vision, Montreal, QC, Canada, pp. 10367-10376, 10-17 October 2021.
[29] Z. Li, P. Xu, X. Chang, L. Yang, Y. Zhang, L. Yao, and X. Chen, "When Object Detection Meets Knowledge Distillation: A Survey," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 45, No. 8, pp. 10555-10579, Aug. 2023.
[30] S. Han, J. Pool, J. Tran, and W. Dally, "Learning both Weights and Connections for Efficient Neural Networks," in Proc. of Conference on Neural Information Processing Systems, Montreal, Canada, pp. 1-9, 07-12 December 2015.
[31] Y. He, J. Lin, Z. Liu, H. Wang, L.-J. Li, and S. Han, "AMC: AutoML for Model Compression and Acceleration on Mobile Devices," in Proc. of European Conference on Computer Vision, Munich, Germany, pp. 815-832, 08-14 September 2018.
[32] S. Pate, and L. Durga, "Survey on Blind-Spot Detection Systems for Improved Vehicle Safety," Propulsion Technology Journal, Vol. 44, No. 5, pp. 2005-2011, Dec. 2023.
[33] W. Kim, H. Yang, and J. Kim, "Blind Spot Detection Radar System Design for Safe Driving of Smart Vehicles," Applied Sciences, Vol. 13, No. 10, pp. 1-23, May. 2023.
[34] S. Wang, Z. Yu, and L. Li, "Detection Model and Correction Method for Quadrant Detector Based Computational Ghost Imaging System," IEEE Sensors Journal, Vol. 24, No. 14, pp. 22565-22574, Jul. 2024.
[35] F. Chollet, "Xception: Deep Learning with Depthwise Separable Convolutions," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, pp. 1800-1807, 21-26 July 2017.
[36] G. Huang, S. Liu, L. v. d. Maaten, and K. Q. Weinberger, "CondenseNet: An Efficient DenseNet Using Learned Group Convolutions," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 2752-2761, 18-23 June 2018.
[37] J. Hu, L. Shen, and G. Sun, "Squeeze-and-Excitation Networks," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 7132-7141, 18-23 June 2018.
[38] Z. Zhang, "Flexible Camera Calibration by Viewing a Plane from Unknown Orientations," in Proc. of IEEE International Conference on Computer Vision, Kerkyra, Greece, pp. 666-673, 20-27 September 1999.
[39] Z. Zhang, "A Flexible New Technique for Camera Calibration," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 22, No. 11, pp. 1330-1334, Nov. 2000.
[40] G. P. Stein, O. Mano, and A. Shashua, "Vision-Based ACC with a Single Camera: Bounds on Range and Range Rate Accuracy," in Proc. of IEEE Conference on Intelligent Vehicles Symposium, Columbus, OH, USA, pp. 120-125, 09-11 June 2003.
[41] Z. Q. XU, and Y. C. Wu, "Application of Data Enhancement Lightweight Neural Network Model in Vehicle Blind Spot Detection," in Proc. of Symposium on Digital Life Technologies, Taichung, Taiwan, pp. 1159-1165, 17-18 May 2024.