以Faster R-CNN模式為基礎之漸進式物件偵測技術

簡易檢索 / 詳目顯示

回結果列表

研究生：	陳延駿 Chen, Yan-Chun.
論文名稱：	以Faster R-CNN模式為基礎之漸進式物件偵測技術 Incremental Object Detection from Moving Vehicles based on Faster R-CNN
指導教授：	王家祥 Wang, Jia-Shung
口試委員:	黃俊堯 Huang, Jun-Yao 陳弘軒 Chen, Hong-Xuan
學位類別：	碩士 Master
系所名稱：
論文出版年：	2019
畢業學年度：	107
語文別：	英文
論文頁數：	18
中文關鍵詞：	影像增強、物件偵測、自動車駕駛
外文關鍵詞：	image reinforcement
相關次數：	點閱：1 下載：0
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

這篇論文將漸進式影像增強的技術運用在漸進圖片中。由於前景的遠處小物體較為模糊，在圖片中占的面積比例也很微小，容易被原始偵測物體的演算法所忽略，為了解決這個問題，我們運用影像增強的概念，使得前景的遠處小物體的辨識率得以提升。
我們使用自駕車拍攝的影片來當作實驗圖片，因此可以使用漸進連續的圖片特性來增進影像剖析的準確率。以Faster R-CNN的模型做為基礎，系統會自動追蹤前後張圖片的物體，並利用歷史紀錄去判定所出現的物體種類是否一致，降低了拿框錯或框的位置不佳的圖片做影像增強的可能性。使用這個系統可以判定在圖片中的遠處是否有物體存在，並提早開始做追蹤與偵測。
這個系統運用了漸進圖片的特性來提早偵測和追蹤前景的遠處小物體，未來希望可以應用在自動車駕駛的領域上，提早可以對較遠的物體做偵測與追蹤。我們的實驗使用自駕車拍攝的現實場景圖片，在車速不快的自駕車上可以提早半秒去偵測遠處的物體。

This thesis presents an incremental object detection system based on Faster R-CNN and image reinforcement. Some small objects in image are too far often occupy only a small portion of the pixels in an image that is hard to detect. To solve this problem, we utilize the concept of image reinforcement to improve us to detect these objects.
Our experimental images are continuously from a moving vehicle. Thus, the information of progressive images can be utilized to improve the accuracy of detection. Our system based on Faster R-CNN model. By using information of progressive images, we tracking the same object in progressive images. Check the history information of the object to prevent we tracking the wrong object and using the wrong information to reinforce the small object detection. Using our system can early detect small objects in the distance.
This novel idea can hopefully be applied on autonomous car driving in the near future. We use incremental images which photo by autonomous car in real world as our dataset. In the dataset which is 25 frame per second, we success to detect the object half of second in advance.

口試委員會審定書    #
致謝    I
中文摘要    II
ABSTRACT    III
CONTENTS    IV
LIST OF FIGURES    VI
Chapter 1.    Introduction    1
Chapter 2.    Related Works    4
2.1    Hand-engineered features and early neural networks    4
2.2    Deep Networks for Object Detection    5
2.2.1    Object Proposals    5
Chapter 3.    Proposed Methods    6
3.1    Basic flow chart    6
3.2    Faster R-CNN Network    7
3.3    Proposal matching    8
3.4    Temporal class decision    9
3.5    Incremental improvement    10
3.6    Early detection decision    11
Chapter 4.    Experimental Results    12
Chapter 5.    Conclusion and Future Work    17
REFERENCES    18
                                

[1] D. G. Lowe. “Distinctive image features from scale-invariant keypoints”. IJCV, 2004
[2] N. Dalal and B. Triggs. “Histograms of oriented gradients for human detection.” In CVPR, 2005
[3] P. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y. LeCun, “Overfeat: Integrated recognition, localization and detection using convolutional networks,” in International Conference on Learning Representations (ICLR), 2014
[4] R. Girshick, J. Donahue, T. Darrell, and J. Malik. “Rich feature hierarchies for accurate object detection and semantic segmentation.” In CVPR, 2014.
[5] R. Girshick. “Fast R-CNN.” In ICCV, 2015.
[6] J. R. Uijlings, K. E. van de Sande, T. Gevers, and A. W. Smeulders, “Selective search for object recognition,” International Journal of Computer Vision (IJCV), 2013.

簡易檢索 / 詳目顯示

相關論文