跳到主要內容

臺灣博碩士論文加值系統

(44.200.194.255) 您好!臺灣時間:2024/07/16 11:21
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

: 
twitterline
研究生:許心平
研究生(外文):SHIU, HSIN-PING
論文名稱:一個用於前方偵測車輛之車距校正方法
論文名稱(外文):A Vehicle Distance Calibration Method for front Vehicle Detection
指導教授:蘇志文蘇志文引用關係
指導教授(外文):SU, CHIN-WEN
口試委員:朱守禮林學億
口試委員(外文):CHU, SLO-LILIN, HSUEH-YI
口試日期:2023-07-18
學位類別:碩士
校院名稱:中原大學
系所名稱:資訊工程學系
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2023
畢業學年度:111
語文別:中文
論文頁數:43
中文關鍵詞:深度學習自動駕駛距離校正物件偵測
外文關鍵詞:Deep LearningAutonomous DrivingDistance CorrectionObject Detection
DOI:10.6840/cycu202301357
相關次數:
  • 被引用被引用:0
  • 點閱點閱:65
  • 評分評分:
  • 下載下載:13
  • 收藏至我的研究室書目清單書目收藏:0
本研究旨在開發一種用於前車三維偵測位置的距離校正方法,以提高自動駕駛系統的精度和可靠性。在現代汽車智能駕駛系統中,前車的三維位置資訊是實現自動駕駛的關鍵因素之一。然而,由於許多因素的影響,基於攝影機畫面測量到的前車位置資訊可能會出現偏差。因此,為確保行駛的安全性,進行前車位置校正是實現可靠自動駕駛的必要步驟之一。本研究的貢獻在於針對前車距離提出了一種新的校正方法,為實現自動駕駛技術的進一步優化提供了可行的想法與初步驗證。本研究結合視覺感知和深度學習技術,分析收集到的光達和影像資料並進行比對,將影像上偵測出的車輛距離作進一步的校正,以補償前車因訓練學習造成的位置偏差,從而提高距離測量的精度和可靠性。具體實驗結果顯示,本研究開發的方法可以在不同道路環境和天氣條件下實現更高精度的前車位置偵測,並且在不同場景下具有良好的通用性。
This study aims to develop a distance correction method for the three-dimensional detection of the position of the vehicle ahead, in order to enhance the accuracy and reliability of autonomous driving systems. In modern automotive intelligent driving systems, the three-dimensional position information of the vehicle ahead is one of the key elements to achieving autonomous driving. However, due to various factors, the position information of the vehicle ahead measured from the camera view may deviate. Therefore, to ensure the safety of driving, correcting the position of the vehicle ahead is one of the essential steps to realize reliable autonomous driving. The contribution of this study is to propose a new correction method for the distance to the vehicle ahead, providing feasible ideas and preliminary verification for further optimization of autonomous driving technology. This research combines visual perception and deep learning technologies to analyze and compare the collected LiDAR and image data, and further corrects the vehicle distance detected on the image to compensate for the position deviation caused by training learning, thereby improving the accuracy and reliability of distance measurement. The specific experimental results show that the method developed in this study can achieve more accurate detection of the position of the vehicle ahead under different road environments and weather conditions, and it has good universality in different scenarios.
目錄
摘要 ................................................................................................................................ I
Abstract ......................................................................................................................... II
致謝詞 ..........................................................................................................................III
目錄 ............................................................................................................................. IV
圖目錄 ......................................................................................................................... VI
表目錄 ....................................................................................................................... VIII
第一章 緒論 ..................................................................................................................1
1.1 研究動機 .............................................................................................................1
1.2 論文架構 .............................................................................................................2
第二章 相關文獻 ..........................................................................................................3
2.1 交通物體檢測 .....................................................................................................3
2.1.1 二階段(Two-stage)物體檢測 .......................................................................3
2.1.2 單階段(One-stage)物體檢測 ........................................................................4
2.2 三維物體距離偵測 .............................................................................................5
第三章 研究方法 ..........................................................................................................9
3.1 物體偵測 .............................................................................................................9
3.2 深度偵測 ...........................................................................................................11
3.3 深度值校正 .......................................................................................................13
3.4 基於YOLOX的多任務模型 ...........................................................................16
第四章 實驗方法 ........................................................................................................19
4.1 實驗環境 ...........................................................................................................19
4.2 實驗資料 ...........................................................................................................19
4.3 實驗結果 ...........................................................................................................22
第五章 結論與未來方向 ............................................................................................33
參考文獻 ......................................................................................................................34
圖目錄
圖 2-1 RCNN 流程圖[3]。…………………………..………………………………………………. 4
圖 2-2 YOLO 流程圖[7]。 .................................................................................................................... 5
圖 2-3 Stereo R-CNN 架構圖[10]。...................................................................................................... 7
圖 2-4 Multi-view 3D Object Detection Network (MV3D)網路架構圖[11]。 ..................................... 7
圖 2-5 Pseudo-LiDAR 方法架構圖[12]。 ............................................................................................ 8
圖 3-1 本方法訓練及檢測流程圖。 ..................................................................................................... 9
圖 3-2 傳統YOLO耦合頭架構與YOLOX解耦頭架構[1]。 ......................................................... 11
圖 3-3 YOLOX與其他精簡模型的實驗結果比較[1]。 .................................................................... 11
圖 3-4 解耦頭與耦合頭的收斂速度比較[1]。 .................................................................................. 12
圖 3-5 傳統 YOLO、YOLOX[1]以及本論文的預測層比較。 ....................................................... 12
圖 3-6 預測邊界框的參數。 ............................................................................................................... 13
圖 3-7 真實光達深度值偵測資料(圖左)與估計深度值偵測(圖右)。.............................................. 14
圖3-8 曲面擬合示意圖。 .................................................................................................................... 15
圖3-9 統計資料經平均數與中位數後之曲面擬合圖形。 ................................................................ 16
圖 3-10 二次多項式曲面擬合示意圖。 ............................................................................................. 16
圖 3-11 本方法的網路結構圖。 ......................................................................................................... 17
圖 4-1 KITTI之物體分類統計[15]。 .................................................................................................. 20
圖 4-2 nuScenes之物體分類統計[16]。 ............................................................................................. 20
圖 4-3 kitti採集數據車輛示意圖[15]。 .............................................................................................. 21
圖 4-4 nuScenes採集數據車輛示意圖[16]。 ..................................................................................... 21
圖 4-5 三維點雲示意圖[17]。 ............................................................................................................ 21
圖 4-6 KITTI(上圖)與nuScenes (下圖)物體邊界框的標註。 ........................................................... 22
圖 4-7 KITTI(上圖)與nuScenes (下圖)深度值的標註。 ................................................................... 22
圖 4-8 KITTI資料集白天場景之偵測結果 ........................................................................................ 25
圖 4-9 nuScenes白天場景之偵測結果。 ............................................................................................ 25
圖 4-10 nuScenes夜晚場景之偵測結果。 .......................................................................................... 26
圖 4-11 原始光達資料(圖左)與本研究之預測深度值偵測結果(圖右)比較。................................ 26
圖 4-12 基於表 4-3產生之曲面擬合圖形。..................................................................................... 29
圖 4-13 基於表 4-5 產生之中為數與平均值之曲面擬合灰階圖形與兩者比較圖。.................... 30
圖 4-14 左為原始資料,中為此模型偵測結果,右為經曲面擬合後的偵測結果 ......................... 32
表目錄
表 4-1 YOLOX四種不同網路深度與寬度比例[1]。 ........................................................................ 23
表 4-2 論文[20]之研究結果。 ............................................................................................................ 23
表 4-3 本研究於GPU RTX4090上之實驗結果。 ............................................................................ 24
表 4-4 為模型使用不同置信度的Loss Function之比較結果 .......................................................... 24
表 4-5 不同資料集測試深度值誤差比較表。 ................................................................................... 24
表 4-6 原始LiDAR與預測深度值誤差統計表。 ............................................................................. 28
表 4-7 基於座標軸X之原始LiDAR與預測深度值誤差統計表。 ................................................ 28
表 4-8 基於座標軸X之原始LiDAR與預測深度值誤差中位數以及平均值統計表。 ................ 29
表 4-9 基於圖 4-13所產生之曲面擬合方程式................................................................................. 31
表 4-10 深度值誤差校正前與校正後之比較。 ................................................................................. 32
[1] Ge, Z., Liu, S., Wang, F., Li, Z., & Sun, J. (2021). YOLOX: Exceeding YOLO Series in 2021. ArXiv, abs/2107.08430.
[2] Redmon, J., & Farhadi, A. (2018). YOLOv3: An Incremental Improvement. ArXiv, abs/1804.02767.
[3] Girshick, R.B., Donahue, J., Darrell, T., & Malik, J. (2014). Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. 2014 IEEE Conference on Computer Vision and Pattern Recognition, pages 580-587.
[4] Uijlings, J.R., Sande, K.E., Gevers, T., & Smeulders, A.W. (2013). Selective Search for Object Recognition. International Journal of Computer Vision, 104, pages 154- 171.
[5] Girshick, R.B. (2015). Fast R-CNN. 2015 IEEE International Conference on Computer Vision (ICCV), pages 1440-1448.
[6] Ren, S., He, K., Girshick, R.B., & Sun, J. (2015). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, pages 1137-1149.
[7] Redmon, J., Divvala, S.K., Girshick, R.B., & Farhadi, A. (2016). You Only Look Once: Unified, Real-Time Object Detection. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 779-788.
[8] Redmon, J., & Farhadi, A. (2017). YOLO9000: Better, Faster, Stronger. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6517-6525.
[9] Lin, T., Dollár, P., Girshick, R.B., He, K., Hariharan, B., & Belongie, S.J. (2017). Feature Pyramid Networks for Object Detection. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 936-944.
[10] Li, P., Chen, X., & Shen, S. (2019). Stereo R-CNN Based 3D Object Detection for Autonomous Driving. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7636-7644.
[11] Chen, X., Ma, H., Wan, J., Li, B., & Xia, T. (2016). Multi-view 3D Object Detection Network for Autonomous Driving. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6526-6534.
[12] Wang, Y., Chao, W., Garg, D., Hariharan, B., Campbell, M.E., & Weinberger, K.Q. (2018). Pseudo-LiDAR From Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8437-8445.
[13] Glenn Jocher et al. yolov5. https://github.com/ultralytics/yolov5, 2021.
[14] Bochkovskiy, A., Wang, C., & Liao, H.M. (2020). YOLOv4: Optimal Speed and Accuracy of Object Detection. ArXiv, abs/2004.10934.
[15] Geiger, A., Lenz, P., Stiller, C., & Urtasun, R. (2013). Vision meets robotics: The KITTI dataset. The International Journal of Robotics Research, 32, pages 1231-1237.
[16] Caesar, H., Bankiti, V., Lang, A.H., Vora, S., Liong, V.E., Xu, Q., Krishnan, A., Pan, Y., Baldan, G., & Beijbom, O. (2019). nuScenes: A Multimodal Dataset for Autonomous Driving. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 11618-11628.
[17] Kui Xu. kitti_object_vis. https://github.com/kuixu/kitti_object_vis, 2021.
[18] Lin, T., Goyal, P., Girshick, R.B., He, K., & Dollár, P. (2020). Focal Loss for Dense Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42, pages 318-327.
[19] Liu, S., Qi, L., Qin, H., Shi, J., & Jia, J. (2018). Path Aggregation Network for Instance Segmentation. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8759-8768.
[20] 莊凱丞(2022)。一個用於偵測可行駛區域與前車三維位置的多任務學習框架。﹝碩士論文。私立中原大學﹞臺灣博碩士論文知識加值系統。 https://hdl.handle.net/11296/5rjh7x。
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊