影像暨距離資訊感知融合之低照度行人偵測系統

Pedestrian Detection System in Low Illumination Conditions through Data Fusion of Image and Range Sensor

黃邦庭(Pang-Ting Huang)

指導教授 ： 傅立成

共同指導教授 ： 蕭培墉(Pei-Yung Hsiao)

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2014年)

https://doi.org/10.6342/NTU.2014.01328

摘要

近幾年以來，智慧車輛的發展越趨快速，智慧車輛結合數種不同的技術，包含了對於駕駛者的輔助，甚至是自動化的駕駛服務。在這些功能的實現上，都倚賴系統對於環境的感知，其中又分為對於號誌、信號等的辨識，以及對於路況的偵測。在路況偵測的部分，對於路上障礙物的偵測更是重要的一環，而障礙物的種類，無非是路上的車輛以及行人，其中行人更是路上障礙物中最脆弱也最需要被偵測的對象。
行人偵測在近二十年來，主要著重在影像方面的辨識，然而考慮到其應用的範疇，在不同的環境下仍有它的限制。其中在日出將至及即將入夜時的時段，由於路況亮度不佳，加以車流量大等原因，更是事故頻傳的時段。在這樣的照度狀態下，影像特徵容易被環境的狀況所影響，造成其效果不彰。
為了處理這樣的狀況，本論文提出了影像暨距離資訊感知融合之低照度行人偵測系統。在影像方面利用結合方向梯度直方圖以及對數加權樣式等兩特徵的結合，配合動態亮度偵測器，在影像方面對低照度環境改良；在距離感知融合方面，在分析觀察之結果之後，利用分析兩感知器的機率模型建構出兩訊號源之關係，並藉由兩感知器之特點，修正原本之偵測結果，進一步得到在低照度環境下依然有穩定的偵測效果並保持實時運算的行人偵測系統。

關鍵字

行人偵測 ； 低照度 ； 距離資訊 ； 感知融合

並列摘要

The development of pedestrian detection techniques mainly focused on the research on finding suitable visual feature in the past decades. However, illumination is not only important to enable human visual ability to view the surroundings, but is also very critical to the choice of visual features in the vision-based detection methods. Since the visual information can be greatly affected by different illuminations, the resulting detection methods can produce unreliable results. To cope with such difficult situation due to uncontrolled lighting conditions, an Image-Range Fusion System (IRFS) is proposed by applying the image data from a camera and the range data from a range sensor simultaneously. For the image part, a Dynamically Illuminated Object (DIO) detector is proposed to overcome the possible problem caused by the uncertain partial lighting condition within a low-illumination environment. Specifically, the DIO detector applies two kinds of features including the Histograms of Oriented Gradients (HOG) for representing the shape information and the Logarithm Weighted Pattern (LWP) for the textural information. Note that LWP gives an 86% recall rate at 〖10〗^(-4) false positive per window, and range data are sampled under a probabilistic model to reinforce the performance by the precise locating ability of the range sensor. To validate our results, several experiments have been conducted, and the overall system performance is shown to be 92.51% / 88.92% of recall/ precision under real-time computing setting.

並列關鍵字

Pedestrian detection ； low illuminated environment ； range sensor ； sensor fusion

參考文獻

[2] N. Dalal and B. Triggs, "Histograms of Oriented Gradients for Human Detection," in IEEE Conference on Computer Vision and Pattern Recognition, 2005, pp. 886-893 vol. 1.

[3] N. Dalal and B. Triggs, "Histograms of oriented gradients for human detection," in IEEE Conf. on Computer Vision and Pattern Recognition 2005, pp. 886-893.

[4] Q. Zhu, M.-C. Yeh, K.-T. Cheng, and S. Avidan, "Fast human detection using a cascade of histograms of oriented gradients," in IEEE Conf. on Computer Vision and Pattern Recognition, 2006, pp. 1491-1498.

[5] S. Maji, A. C. Berg, and J. Malik, "Classification using intersection kernel support vector machines is efficient," in IEEE Conf. on Computer Vision and Pattern Recognition, 2008, pp. 1-8.

[6] O. Tuzel, F. Porikli, and P. Meer, "Human detection via classification on riemannian manifolds," in IEEE Conf. on Computer Vision and Pattern Recognition, 2007, pp. 1-8.