行人偵測系統發展至今出現了許多優秀的方法，而駕駛輔助系統的普及和自動駕駛汽車的出現更是讓行人偵測有了更高的實用價值與更多的應用空間。由於近年深度學習的興起，漸漸地出現了結合深度學習與行人偵測的研究，但深度學習不論是學習或是偵測時皆需要高階硬體以供其龐大的運算量，阻礙了於行人偵測上的實用性。本研究在不使用並行運算與能在一般硬體運行的條件下，設計出一套應用於車輛影像的行人偵測系統。
在本研究中，我們先根據影片的攝影機狀況預測感興趣區域（ROIs，Region of Interest），以減少不必要的特徵計算與目標搜尋。接著在計算特徵時使用快速特徵金字塔（Fast Feature Pyramids）算法，進一步減少特徵計算階段的耗時。最後以 Cascade DPM（Deformable Part Models）方法偵測出行人。在小幅降低精度（Precision）與召回率（Recall）的狀況下，將整體系統之運算速度提升到Cascade DPM的2.54倍。

There are many mature pedestrian detection methods that had been developed so far. The widespread popularity of driving support system and the emerging of unmanned vehicles let pedestrian detection possesses more practical value and wider application space. Due to the arising of deep learning recently, there is a trend by incorporating deep learning into pedestrian detection. However, deep learning requires high-level hardware and tremendous amount of computation no matter in learning or detection to hinder the practicality of pedestrian detection. In this thesis, a pedestrian detection system is designed for vehicle images without using concurrent computation which can run under general hardware.
In our work, the ROIs (Region of Interest) are firstly predicted based on the camera status of video to reduce unnecessary feature calculation and target search. Then, the Fast Feature Pyramids algorithm is employed to calculate features to further reduce the time spent in the feature calculation phase. Finally, Cascade DPM (Deformable Part Models) method is utilized to detect pedestrians. The speed of our proposed system can uplift the speed to 2.54 times faster than Cascade DPM with slightly lowering precision and recall rate.

[1] (2016-2017). Retrieved from PILOT: http://www.pilotlab.co/
[2] A. Mohan, C. Papageorgiou, and T. Poggio. (2010, April). Example-based object detection in images by components. Pattern Analysis and Machine Intelligence (PAMI), pp. 349-361.
[3] C. Papageorgiou and T. Poggio. (2000). A trainable system for object detection. International Journal of Computer Vision (IJCV), (pp. 15-33).
[4] Charles Dubout and François Fleuret. (2012). Exact Acceleration of Linear Object Detectors. European conference on Computer Vision (ECCV) (pp. 301-311). Florence, Italy: Springer.
[5] Chi-Hong Kuo. (2011). Monocular-vision pedestrian detection and tracking. Unpublished Master Thesis.
[6] D.L. Ruderman and W. Bialek. (1994, Aug.). Statistics of Natural Images: Scaling. Physical Rev. Letters, pp. 814-817.
[7] Dalal, N. (n.d.). INRIA Person Dataset. Retrieved from INRIA: http://pascal.inrialpes.fr/data/human/
[8] Felzenszwalb, P. F. (2012). voc-release5. Retrieved from Discriminatively trained deformable part models: https://dl.dropboxusercontent.com/s/gh7reh931y1wqmf/voc-release5.tgz?dl=0
[9] Felzenszwalb, P. F. (2013). voc-release3.1. Retrieved from Discriminatively trained deformable part models: http://cs.brown.edu/~pff/latent-release3/voc-release3.1.tgz
[10] Junjie Yan, Zhen Lei, Longyin Wen, and Stan Z. Li. (2014). The Fastest Deformable Part Model for Object Detection. Computer Vision and Pattern Recognition (CVPR) (pp. 2497-2504). Columbus, OH, USA: IEEE.
[11] Marco Pedersoli, Andrea Vedaldi, and Jordi Gonzàlez. (2011). A coarse-to-fine approach for fast deformable object detection. Computer Vision and Pattern Recognition (CVPR) (pp. 1353-1360). Colorado Springs, CO, USA, USA: IEEE.
[12] Mohammad Amin Sadeghi, and David Forsyth. (2014). 30Hz Object Detection with DPM V5. European Conference on Computer Vision (ECCV) (pp. 65-79). Zurich, Switzerland: Springer.
[13] N. Dalal and B. Triggs. (2005). Histograms of oriented gradients for human detection. Computer Vision and Pattern Recognition (CVPR) (pp. 886-893). San Diego, CA, USA, USA: IEEE.
[14] Open Source Computer Vision Library. (2017). Retrieved from OpenCV: http://opencv.org/
[15] P. Dollár, S. Belongie, and P. Perona. (2010). The Fastest Pedestrian Detector in the West. British Machine Vision Conf (BMVC).
[16] P. Dollár, Z. Tu, P. Perona, and S. Belongie. (2009). Integral Channel Features. British Machine Vision Conf. (BMVC).
[17] Pedro F. Felzenszwalb, Ross B. Girshick, and David McAllester. (2010). Cascade Object Detection with Deformable Part Models. Computer Vision and Pattern Recognition (CVPR) (pp. 2241-2248). San Francisco, CA, USA: IEEE.
[18] Pedro F. Felzenszwalb, Ross B. Girshick, David McAllester, and Deva Ramanan. (2010). Object Detection with Discriminatively Trained Part-Based Models. IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1627 - 1645.
[19] Pedro Felzenszwalb, David McAllester, and Deva Ramanan. (2008). A Discriminatively Trained, Multiscale, Deformable Part Model. Computer Vision and Pattern Recognition (CVPR). Anchorage, AK, USA: IEEE.
[20] Perspective (graphical). (n.d.). Retrieved from Wikipedia: https://en.wikipedia.org/wiki/Perspective_(graphical)#/media/File:Drawing_Square_in_Perspective_2.svg
[21] Piotr Dollár, Ron Appel, Serge Belongie, and Pietro Perona. (2014, January 16). Fast Feature Pyramids for Object Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1532-1545.
[22] Ruderman, D. (1994). The Statistics of Natural Images. Network: Computation, pp. 517-548.
[23] S. Belongie, J. Malik, and J. Puzicha. (2001). Matching shapes. International Conference on Computer Vision (ICCV) (pp. 454-461). Vancouver, Canada: IEEE.
[24] The PASCAL Visual Object Classes Homepage. (n.d.). Retrieved from PASCAL VOC project: http://host.robots.ox.ac.uk/pascal/VOC/
[25] Y. Ke and R. Sukthankar. (2004). Pca-sift: A more distinctive representation. Computer Vision and Pattern Recognition (CVPR) (pp. 66-75). Washington, DC, USA, USA: IEEE.