結合距離資訊的方向梯度直方圖行人偵測法

Range assisted HOG pedestrian detection

周大為(Ta-Wei Chou)

指導教授 ： 王傑智

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.00189

摘要

我們提出將從雷射測距儀得到的距離資訊加入影像，從而增進方向梯度直方圖行人偵測法的偵測成效。本文的主要貢獻為：將距離資訊加入行人偵測演算法中，根據距離來計算影像上合適的偵測區塊，並決定恰當的偵測縮放比例 (scale)，從而降低偵測時的誤偵率 (false positive rate)，以及縮小搜尋合適的偵測縮放比例的範圍。偵測成果與目前最佳的行人偵測演算法之一的 「整合通道特徵法 (Integral Channel Features)」 相當，但僅有其十分之一的誤偵率。當以相同誤偵率進行比較時，相較於整合通道特徵法 44.4% 的平均偵測率，結合距離資訊的方向梯度演算法可達到 55.6% 的平均偵測率。另外，相較於整合通道特徵法 53.1% 的平均失誤率，我們的方法可達到 43.8% 的平均失誤率。

關鍵字

行人偵測 ； 感測器整合 ； 整合通道特徵法 ； 方向梯度直方圖法 ； 距離輔助的方向梯度直方圖法

並列摘要

We proposed to fuse information from laser scanner and camera to improve the performance of Histogram of Oriented Gradients (HOG) pedestrian detector. The key contribution is utilizing the range information into the pedestrian detector, so as to reduce the search space of detection scales, and more importantly, reduce the false positive detection. While having one tenth of the false positive per image (FPPI) compared to state-of-the-art Integral Channel Feature (ICF) pedestrian detector, comparable detection rate was achieved. When comparing under the same FPPI, average hit rate at 55.6% is achieved by the proposed range assisted HOG (Ra- HOG), compared to ICF’s 46.9%. Average miss rate at 44.4% is achieved by RaHOG, compared to ICF’s 53.1%.

並列關鍵字

Range assisted HOG ； Histogram of Oriented Gradients ； HOG ； Integral Channel Features ； Sensor fusion ； Pedestrian detection

參考文獻

Badino, H., Huber, D., & Kanade, T. (2011). Integrating lidar into stereo for fast and improved disparity computation. 3D Imaging, Modeling, Processing, Visualization and Transmission, International Conference on, 405 –412.

Belongie, S., Malik, J., & Puzicha, J. (2001). Matching shapes. Computer Vision (ICCV), Proceedings, IEEE International Conference on, 1, 454 –461.

Campbell, J. B. (1996). Introduction to Remote Sensing: Second Edition. Guilford Press. Carlbom, I. & Paciorek, J. (1978). Planar geometric projections and viewing transformations. ACM Computing Surveys, 10(4), 465–502.

Dalal, N. & Triggs, B. (2005). Histograms of oriented gradients for human detection. IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 1, 886–893.

Doll’ar, P.,Wojek, C., Schiele, B., & Perona, P. (2012). Pedestrian detection: An evaluation of the state of the art. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI),34, 743–761.