本論文針對先進駕駛輔助系統(ADAS)中的行人偵測，其重要的特徵擷取提出新的演算法，以目前常被研究的HOG(Histograms of Oriented Gradients)特徵擷取演算法為基礎，提出兩種簡化後的演算法，分別為GC(Gradient Comprison)-HOG和S(Simplified)-HOG，這兩種演算法都只需要簡單的比較和加法，相對HOG演算法而言計算複雜度簡單很多，使得硬體設計上計算單元的面積和延遲下降許多，且對特徵値使用不同的正規化方式，使得硬體所需的儲存空間大幅減少。對於SLBP-HOG和S-HOG提出相對的硬體架構，以降低硬體面積為優先，使用最少的儲存空間和計算單元，設計出低成本的特徵擷取硬體，相較於HOG演算法，硬體面積僅為原來的1/7，工作頻率可以快上25~50%，為了能更明確評估硬體數據，分別跑Standard-Cell Synthesis和FPGA Implementation來跟HOG相關論文比較。而GC-HOG和S-HOG這兩種特徵擷取演算法相對HOG來說準確度是有些微下降的，使用INRIA的Data Set和用SVM去訓練出分類器出來，行人偵測準確度是相差4%左右，然而以DET(Detection Error Tradeoff)曲線來觀察的話GC-HOG和S-HOG是比HOG好的，最後還有拿KITTI提供的街景圖去看看不同特徵擷取演算法在行人偵測上的效果，雖然非行人誤判的情況較多，但是行人誤判的情況跟HOG差不多。

This thesis presents two versions of simplifications, Gradient Comprison HOG (GC-HOG) and Simplified HOG (S-HOG) for Histograms of Oriented Gradients (HOG), a feature extraction algorithm widely used for pedestrian detection in advanced driver assisted systems (ADAS). The simplified HOG only needs simple hardware components, resulting in significant reduction in hardware area cost and delay of arithmetic components. Besides, the size of the internal memory buffers is also reduced due to the simplification in normalization. Compared with the original HOG design, the proposed implementations takes only about 1/7 of total area with 25%~50% speedup of clock rate. The proposed designs are realized using cell-based and FPGA approaches for comparison. When applied to pedestrian detection using INRIA dataset with Support Vector Machine (SVM) for training, the detection accuracy of the proposed GC-HOG and S-HOG designs drops about 4%. But in accuracy metric of Detection Error Tradeoff (DET), the proposed designs even outperforms the original HOG. We also do experiments for pedestrian detection with data from KITTI and observe that the miss detection rate of non-pedestrian with our approaches is slightly increased while the miss rate of pedestrian is almost the same compared with the original HOG.

摘要 i
Abstract ii
目錄 iii
表目錄 v
圖目錄 vi
式目錄 viii
第1章 緒論 1
1.1 本文大鋼 1
1.2 研究動機 1
第2章 研究背景與相關知識 2
2.1 先進駕駛輔助系統(ADAS) 2
2.2 物件偵測 3
2.3 行人偵測 4
2.4 Histogram of Oriented Gradients (HOG) 6
2.5 Evolution of HOG 10
第3章 HOG演算法簡化 12
3.1 簡化HOG演算法介紹 12
3.2 GC-HOG演算法 14
3.2.1 Gradient Comparison 16
3.2.2 Binary Gradient Vote 18
3.2.3 Cell Binarization 18
3.3 S-HOG演算法 20
3.4 小結 22
第4章 簡化HOG硬體設計 24
4.1 GC-HOG(Block Binarization)硬體 24
4.1.1 Gradient Comparison 25
4.1.2 Binary Gradient Vote 27
4.1.3 Block Binarization 31
4.2 GC-HOG(Cell Binarization)硬體 34
4.3 S-HOG硬體 36
4.4 小結 37
第5章 實驗結果及分析 38
5.1 演算法準確度比較 38
5.1.1 GC-HOG演算法 38
5.1.2 S-HOG演算法 40
5.1.3 論文比較 41
5.2 硬體設計數據比較 42
5.2.1 Standard-Cell Synthesis 42
5.2.2 FPGA Implementation 43
5.3 行人偵測上演算法成效比較 45
第6章 結論與未來展望 50
6.1 結論 50
6.2 未來展望 50
參考文獻 51
附錄一 Data Set 55
附錄二 分類器 56

[1] M. Rezaei, M. Terauchi and R. Klette, “Robust Vehicle Detection and Distance Estimation Under Challenging Lighting Conditions” IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 5, pp. 2723 – 2743, Oct. 2015.
[2] C. Guo, et al., “A Multimodal ADAS System for Unmarked Urban Scenarios Based on Road Context Understanding” IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 4, pp. 1690 – 1704, Aug. 2015.
[3] S. Liu, Y. Huang and R. Zhang, “Obstacle Recognition for ADAS Using Stereovision and Snake Models” IEEE 17th International Conference on Intelligent Transportation Systems , pp. 99 –104, Oct. 2014.
[4] C. C. Wong, et al., “A Smart Moving Vehicle Detection System Using Motion Vectors and Generic Line Features” IEEE Transactions on Consumer Electronics, Vol. 61, No. 3, pp. 384 – 392, Aug. 2015
[5] A. D. Jarnea, et al., “Advanced Driver Assistance System for Overtaking Maneuver on a Highway” IEEE 19th International Conference on System Theory, Control and Computing, pp. 759 –764, Oct.2015.
[6] R. O. Chavez-Garcia and O. Aycard, “Multiple Sensor Fusion and Classification for Moving Object Detection and Tracking” IEEE Transactions On Intelligent Transportation Systems, vol. 17, no. 2, pp. 525 – 534, Feb. 2016.
[7] P. Viswanath, et al., “A Diverse Low Cost High Performance Platform for Advanced Driver Assistance System (ADAS) Applications” IEEE Conference on Computer Vision and Pattern Recognition Workshops, pp. 819 – 827, 2016.
[8] M. Cheon, W. Lee, C. Yoon and M. Park, “Vision-Based Vehicle Detection System With Consideration of the Detecting Location,” IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 3, pp. 1243 – 1252, 2012.
[9] C. Zhang, K. H. Chung and J. Kim. “Region-of-interest reduction using edge and depth images for pedestrian detection in urban areas,”SoC Design Conference (ISOCC), 2015 International, IEEE, pp. 161 – 162, 2015.
[10] L. W. Tsai, J. W. Hsieh and K. C. Fan, “Vehicle Detection Using Normalized Color and Edge Map,” IEEE Transactions on Image Processing, vol. 16, no. 3, pp. 850 – 864, 2007.
[11] B.D. Lucas and T. Kanade, “An Iterative Image Registration Technique with an Application to Stereo Vision”, DARPA Image Understanding Workshop, pp. 121 - 130, 1981.
[12] V. D. Nguyen, et al., “Toward Real-Time Vehicle Detection Using Stereo Vision and an Evolutionary Algorithm,” IEEE 75th Vehicular Technology Conference (VTC Spring), pp. 1 – 5, 2012.
[13] Z. Lin and L. S. Davis, “Shape-Based Human Detection and Segmentation via Hierarchical Part-Template Matching,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 32, no. 4, pp. 604 – 618, 2010.
[14] B. Kröse and P. van der Smagt, “An Introduction to Neural Networks Amsterdam,” Netherlands University of Amsterdam, 1996.
[15] V. N. Vapnik, “The Nature of Statistical Learning Theory,” NewYork,NY, USA Springer-Verlag, 1995.
[16] Y. Freund and R. E. Schapire, “A decision-theoretic generalization of on-line learning and an application to boosting,” presented at the 2nd European Conf. Computational Learning Theory, Florham Park, NJ,
[17] T. Ojala, M. Pietikainen and T. Ahonen, “A comparative study of texture measures with classification based on featured distributions,” Pattern Recognition, 29(1): 51~59, 1996.
[18] Y. Mu, et al., “Discriminative local binary patterns for human detection in personal album,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 1 – 8, 2008.
[19] P. Viola and M. Jones, “Robust Real-time Object Detection,” 2nd international workshop on statistical and computational theories of vision - modeling, learning, computing, and sampling, pp. 1-25, 2001.
[20] P. Viola, M. Jones, “Rapid object detection using a boosted cascade of simple features,” Computer Vision and Pattern Recognition, vol. 1, pp. I-511 – I-518 , 2001.
[21] D. G. Lowe, “Distinctive image features from scale-invariant key points,” Int. J.Comp. Vis., vol. 60, no. 2, pp. 91–110, 2004.
[22] N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection. ” Computer Society Conference on Computer Vision and Pattern Recognition, IEEE, vol. 1, pp. 886 – 893, 2005.
[23] X. Wang, T.X. Han, and S. Yan. “An HOG-LBP human detector with partial occlusion handling. ” IEEE 12th International Conference on Computer Vision, pp. 32 – 39, 2009.
[24] Q. Zhu, et al., “Fast Human Detection Using a Cascade of Histograms of Oriented Gradients. ” IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2, pp. 1491 – 1498, 2006.
[25] C. Zhou, et al., “Human Detection Based on Fusion of Histograms of Oriented Gradients and Main Partial Features. ”in 2nd International Congress on Image and Signal Processing, IEEE, pp. 1 – 5, 2009.
[26] P. Y. Chen, et al., “An Efficient Hardware Implementation of HOG Feature Extraction for Human Detection” Transactions on intelligent transportation systems, IEEE, vol. 15, no. 2, 2014.
[27] J. F. Wang, et al. ,“Simplifying HOG Arithmetic for Speedy Hardware Realization, ” Asia Pacific Conference on Circuits and Systems, IEEE, pp. 61 – 64 , 2014.
[28] M. Hemmati et al. “HOG Feature Extractor Hardware Accelerator for Real-time Pedestrian Detection,” 17th Euromicro Conference on Digital System Design, IEEE, pp. 543 – 550, 2014.
[29] 詹鈞貿，影像特徵擷取演算法之硬體加速，國立中山大學資訊工程所碩士論文，民國104年。
[30] A. Khan, et al., “Hardware Architecture and Optimization of Sliding Window Based Pedestrian Detection on FPGA for High Resolution Images by Varying Local Features, ” International Conference on Very Large Scale Integration, IFIP/IEEE, pp. 142 – 148, 2015.
[31] J. Rettkowski, et al., “ Real-time pedestrian detection on a xilinx zynq using the HOG algorithm,” International Conference on ReConFigurable Computing and FPGAs, IEEE, pp. 1 – 8, 2015.
[32] C. Blair, et al., “Characterizing a Heterogeneous System for Person Detection in Video Using Histograms of Oriented Gradients: Power Versus Speed Versus Accuracy,” IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol.3, no. 2, pp. 236 – 247, 2013.
[33] R. Kadota, et al., “Hardware Architecture for HOG Feature Extraction,” Fifth International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IEEE, pp. 1330 – 1333, 2009.
[34] S. Lee, et al., “HOG feature extractor circuit for real-time human and vehicle detection,” TENCON 2012 IEEE Region 10 Conference, IEEE, pp. 1 – 5, 2012.
[35] S.-F. Hsiao, J.-M. Chan and C.-H. Wang, “Hardware design of histograms of oriented gradients based on local binary pattern and binarization,” IEEE Asia Pacific Conference on Circuits and Systems, IEEE, pp. 433 – 435, 2016.