具方向性之空拍車輛偵測方法

Oriented vehicle detection for aerial images

謝侑衽(YOU-JEN HSIEH)

指導教授 ： 蘇志文

中原大學/電機資訊學院/資訊工程研究所/碩士(2020年)

https://doi.org/10.6840/cycu202000523

摘要

為了使交通系統更加智慧，配備高解析度攝影機的無人機已成為觀察交通行為的新工具。與一般公路上架設的交通攝影機相比，使用空拍影片有幾個優點。例如，車輛之間沒有遮蔽問題，車輛形成的軌跡完整，從上空垂直俯瞰下的車輛具備相似輪廓等。因此，利用空拍影片偵測並分類車輛是未來道路交通分析的關鍵之一。近年來，卷積神經網路（CNN）的發展逐漸提高了圖像分類和目標偵測的準確性和強穩性。然而，大多數基於CNN的物體偵測都是透過邊界框來定位每個物體，但邊界框卻無法緊密擬合物體的外型輪廓。因此，我們使用安裝了攝影機的無人機（UAV），從距離地面75至100公尺的高度拍攝影片，並偵測車輛以提取出具方向性的矩形車輛框。本論文提出的方法主要是基於深度學習模型SCRDet進行改良，實驗結果證明我們提出的改進版本，可以更精確地定位出具方向性的矩形車輛框。

關鍵字

空拍影像處理 ； 深度學習 ； 車輛分類 ； 車輛偵測

並列摘要

In order to make the transportation system more intelligent, a drone equipped with high resolution camera has become a novel utility for the observation of traffic behavior. There are several advantages for using aerial video in contrast with on-road traffic camera: such as no occlusion between vehicles, complete trajectory, similar silhouettes at top view. Therefore, the detection and classification of vehicles in aerial video is one of the key issues for the road traffic analysis in future. The advance of Convolution Neural Networks (CNNs) dramatic increases the accuracy and robustness of image classification and object detection. However, most of the CNN-based object detections locate each object by a bounding box which cannot close fitting the contour of object. Instead, we used a drone vehicle and attached a camera to record video at 75 to 100 meters height from the ground, and perform oriented object detection to extract the area of vehicles. Our proposed method is modified based on the deep learning model SCRDet[1]. The experimental result shows that our proposed modified version is more effective in detecting oriented bounding box that fit the vehicle precisely.

並列關鍵字

Aerial image processing ； deep learning ； Vehicle classification ； Vehicle detection

參考文獻

[1] X. Yang, J. Yang, J. Yan, Y. Zhang, T. Zhang, Z. Guo, X. Sun, and K. Fu. “SCRdet: Towards more robust detection for small, cluttered and rotated objects,” In Proceedings of the IEEE International Conference on Computer Vision (ICCV), November 2019.

[2] R. Girshick, J. Donahue, T. Darrell, and J. Malik. “Rich feature hierarchies for accurate object detection and semantic segmentation,” The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2014.

[3] R. Girshick. “Fast R-CNN,” The IEEE International Conference on Computer Vision (ICCV), December 2015.

[4] Y. Xu, G. Yu, and Y. Wang. “Car Detection from Low-Altitude UAV Imagery with the Faster R-CNN,” Journal of Advanced Transportation, pp.40-49, August 2017.

[5] S. Ren, K. He, R. Girshick, and J. Sun. “Faster R-CNN: Towards real-time object detection with region proposal networks,” The IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1137-1149, June 2016.