2017 4th NAFOSTED Conference on Information and Computer Science
Accurate and Low Complex Cell Histogram
Generation By Bypass The Gradient of Pixel
Computation
Huy-Hung Ho, Ngoc-Sinh Nguyen, Duy-Hieu Bui, Xuan-Tu Tran
SIS Laboratory, VNU University of Engineering and Technology (VNU-UET),
144 Xuan Thuy Road, Cau Giay, Hanoi, Vietnam. Email: {sinhnn 55,hieubd,tutx}@vnu.edu.vn
Abstract—Histogram of Oriented Gradient (HOG) is a popular
feature description for the purpose of object detection. However,
HOG algorithm requires a high performance system because of
its complex operation set. In HOG algorithm, the cell histogram
generation is one of the most complex part, it uses inverse
tangent, square, square root, floating point multiplication. In this
paper, we propose an accurate and low complex cell histogram
generation by bypass the gradient of pixel computation. It
employs the bin’s boundary angle method to determine the two
quantized angles. However, instead of choosing an approximate
value of tan, the nearest greater and the nearest smaller of each
tan value from the ratios between pixel’s derivative in y and x
direction are used. The magnitudes of two bins are the solutions
of a system of two equations, which represents the equality of the
gradient of a pixel and its two bins in both vertical and horizontal
direction. The proposed method spends only 30 addition and
40 shift operations to caculate two bins of a pixel. Simulation
results show that the percentage of error when reconstructing the
differences in x and y direction are always less than 2% with 8-
bit length of the fractional part. Additionally, manipulating the
precision of gradient magnitude is very simple by pre-defined
sine and cosine values of quantized angles. The synthesis results
of a hardware implementation of the proposed method occupy
3.57 KGEs in 45nm NanGate standard cell library. The hardware
module runs at the maximum frequency of 400 MHz, and the
throughput is 0.4 pixel/ns for a single module. It is able to support
48 fps with 4K UHD resolution.
I. I NTRODUCTION
The Histogram of Oriented Gradients (HOG) [1] is a feature
descriptor used in computer vision and image processing for
object detection. It gets the quantity of strong characteristics
from the shape change of the object by dividing a local domain
into plural blocks and making the incline of each histogram.
Practically, HOG feature achieves very high accuracy level,
it is up to 96.6% of the detection rate with 20.7% of the
false positive rate [2]. Hence, it is the key role of wide range
application domains including robotic, security surveillance
and automotive. However, the complexity level of HOG is
the most difficult problem when using in embedded systems
or implementing into hardware.
The HOG algorithm can be separated into 3 phases: cell
histogram generation, block normalization, and the SVM
classification. The first phase plays as the most complex one
and consumes the most energy. Because extracting the feature
of each pixel requires series of complex operations: inverse
tangent, square, square root, floating point multiplication.
978-1-5386-3210-9/17/$31.00 ©2017 IEEE 201
Additionally, in the age of cloud computing, the two last
phases can be done in the server for IoT devices. Hence,
simplifying cell histogram generation part is very necessary,
especially for IoT devices.
Many researchers have proposed diverse approximate meth-
ods such as LUT or CORDIC or bin’s boundary angles in
cell histogram generation to reduce complexity levels. Even
though approximate models reduce the accuracy of extracting
pixel feature, they still require large memory areas or use many
iterations.
In this work, we would like to propose a new method,
which extracts features of a pixel accurately and is suitable
for low resource systems and hardware implementation. The
difference in this method is that it is not an approximate model,
and it computes the quantized orientations and corresponding
gradient magnitude of each quantized orientations directly. The
idea comes from the properties of decomposing the gradient
of a pixel into two components (bins). The proposed method
includes only addition and shift operation. As our simulation
results, extracting non-normalization features of a pixel spends
about 30 additions and 40 shifts for 8-bit of fraction of
magnitude. In the term of accuracy, error when reconstructing
the pixel’s derivatives in both direction from voted bins is
always less than 2%. A hardware implementation occupies
about 3.57 KGEs in NanGate 45nm standard cell library. The
maximum operating frequency of the hardware is 400MHz,
and its throughput is up to 0.4 pixel/ns.
The rests of the paper are organized as follows. Section II
introduces the cell histogram generation and some previous
optimization works. Section III shows the detail of proposed
method. Section IV presents details of the hardware imple-
mentation of proposed method and simulation results. Finally,
the conclusion and the perspectives are presented in the section
V.
II. C ELL HISTOGRAM GENERATION AND ITS
APPROXIMATIONS
A. Conventional cell histogram generation
Histogram of Oriented Gradient (HOG), pioneered by
DALAL and TRIGGS [1], becomes one of the most popular
methods for feature extraction. In the conventional HOG, the
cell histogram generation part is the most dominant power
consumption, up to 58% HOG power in Takagi’s work [3].
 2017 4th NAFOSTED Conference on Information and Computer Science

Authors in [4] analyzed the cell histogram generation part ac-

count for 91% workload in detection window-based approach. −−−−−→
||M (x, y)||≈ |dx | + |dy | (8)
Equation 1 - 6 present all equations using in the conven- −−−−−→
tional HOG. As shown in these equations, the high complexity ||M (x, y)||≈ max((0.875 × gmax + 0.5gmin ), gmax ) (9)
of the cell histogram generation comes from its complex −−−−−→ gmin
||M (x, y)||≈ |dx | + |dy | − (10)
operations: two squares, one square root, one inverse tangent, 2
four floating point multiplications. where gmax and gmin is the maximum and minimum of
Pixel derivatives with respect to x,y: |dx |, |dy |, respectively.

dx = I(x, y + 1) − I(x, y − 1) (1) C. Gradient orientation approximations

dy = I(x + 1, y) − I(x − 1, y) (2) Inverse tangent operation is the most challenging operation
for hardware implementation. There are several approaches
Gradient magnitude calculation: in practice to implement this non-linear operation. Iandola
[6] used full LUT for inverse tangent, and consumed 256KB
−−−−−→ q
||M (x, y)|| = d2x + d2y (3) memory. Peker [11] combined the LUT and the piece-wise
approximations to obtain higher accuracy with less demand
Gradient orientation/angle calculation: on the resources. Takaghi [3] applied by CORDIC to avoid
d  high memory resources. However, CORDIC module is imple-
y mented with many iterations to obtain acceptable calculation
α = arctan (4)
dx precision.
Identifying two voted bins: D. Determining two quantized orientation without gradient
orientation
θi < α < θi+1 (5)
−→ −−−−−→ θj+1 − α A more effective approach of extracting HOG features of a
||Bθi || = ||M (x, y)|| × (6) pixel is bin’s boundary angles. In this method, the actual angle
w
−−−→ −−−−−→ α − θi of the gradient does not need to be calculated. It uses Equation
||Bθi+1 ||= ||M (x, y)|| × (7) 11 to determine the two quantized angles. In order to simplify
w
the implementation, Equation 11 has been converted to other
where formulas. Chen [9] proposed Equation 12 with fixed tan θ. In
−−−−−→ contrast, publications [2], [3], [7], [8], [12], [13] multiplied
• ||M (x, y)|| is the gradient magnitude of pixel I(x, y);
• α is the gradient orientation of pixel I(x, y); tan θ with a constant A, then rounding A × tan θ value as
• θj , θj+1 are two quantized orientations; Equation 13.
−→ −−−→
• ||Bθi ||, ||Bθi+1 || are the magnitude of two quantized
orientations: θi , θj+1 ; dy
tan θ1 < < tan θ2 (11)
• w is the angle distance between two continuous orienta- dx

tions, w = 20. tan θ1 × dxP< dy < tan θ2 × dx
n (12)
In order to reduce the complexity of cell histogram gen- tan θ = i=−m 2i , m, n ∈ N
eration and to be able for hardware implementation, several bA × tan θ1 e × dx < A × dy < bA × tan θ2 e × dx (13)
approximate algorithms have been applied. All of approximate
Converting Equation 11 into Equation 12 and 13 has caution
models have been employed to avoid non-linear operations in
of choosing fixed tan θ or the A values. Those constants
gradient magnitude and gradient orientation computation.
affects to determine the two quantized orientation directly. In
fact, Equation 12 cannot present a rational number, whose
B. Gradient magnitude approximations d
divisor is a prime number. Let’s take an example, a ( dxy ≡
56
Computing gradient magnitude by L2-norm (Equation 3) 97 = 0.57732 < tan 30) expects to contribute its gradient
97
spends two squares and one square root, which are inefficient to two angles: 10 and 30 degree, but the 168 = 0.577381 is
to be implemented into hardware To solve this problem, the in the angle range (30:50) degree. In this case, the A value
simplest method is using LUT as used in Kadota’s work [5]. have to be equal or greater than 105 to distinct the those two
Because the LUT method requires very high areas cost, so values. However, increasing the A values means increasing the
researchers have applied approximate computation models. complexity of implementation.
Iandola [6] and Suleiman [7] employed L1-norm as Equation Another disadvantage of this method is that it does not
8 instead of L2-norm. Hsiao and Chen [8], [9] used an ap- have enough information for voting magnitude as Equation
proximate model, called square root approximation technique 6 and 7. In this case, those mentioned papers using this
(SRA) as Equation 9. Munteanu [10] implemented Equation method solved this problem by implementing fixed weight for
10. However, those approximate models reduce the HOG voting magnitude, usually equal distribution. Consequently, all
accuracy level. gradients have the same magnitudes and the same orientation

978-1-5386-3210-9/17/$31.00 ©2017 IEEE 202

 2017 4th NAFOSTED Conference on Information and Computer Science

range, but they have different orientations. They will have the
𝒚
same HOG features. 𝑴
III. P ROPOSED CELL HISTOGRAM GENERATION
In this work, we proposed a robust and low complexity 𝑩𝟐 × 𝐬𝐢𝐧 𝜽𝒊+𝟏 𝑩𝟐
method to identify two feature bins of a pixel. This method
has no calculation of the gradient of pixel. To determine the 𝒅𝒚
two quantized orientations exactly, a trusted form of Equation
d
11 is proposed. It is from the limitation of the dxy and the
tan θ values. Gradient magnitudes of two bins are solutions
of a system of two equations instead of voting gradient. 𝑩𝟏
𝑩𝟏 × 𝐬𝐢𝐧 𝜽𝒊 𝜽𝒊+𝟏
𝜽𝒊
A. Robust quantized angles determination
Equation 14 shows our basic idea to calculate the quantized 𝑶 𝒙
𝒅𝒙
angles. In stead of choosing fixed bit length of fractional
P iof tan(θ) values or representing a tan(θ) value by sum
part
𝑩𝟐 × 𝐜𝐨𝐬 𝜽𝒊+𝟏
2 , our proposed method limits each tan(θ) by two rational 𝑩𝟏 × 𝐜𝐨𝐬 𝜽𝒊
d As
numbers: its nearest smaller dxy , called Bs
and its nearest
dy Ag
greater dx , called Bg . Those fractions do not only depend
dy
on the tan values, it is from examining all cases of dx and Fig. 1. Decomposition a vector into form of two vectors
tan values to reach the expected results.
As Ag Ag d
< tan(θ) < (14) • B
g
is the nearest greater or equal number dxy of tan(θ)
Bs Bg value;
As A • As , Bs , Ag , Bg is integer number in |dx |, |dy | range.
Table I shows all B and Bgg in a case of 8-bit length pixel.
s • θ is quantized orientation;
With those rational numbers in the table, multiplying with
• α is the gradient orientation of the current pixel.
very large number such as 105 is not necessary to determine
quantized orientation exactly. B. Magnitude of two bins calculation
As the discussion in section II-D, the results of bin’s bound-
TABLE I ary angles do not provide enough arguments for voting. Our
dy
tan θ AND ITS NEAREST VALUES d
x work proposed a system of two equations from information:
θi , θi+1 , dx and dy to calculate gradient magnitude of two bins
Nearest smaller Nearest greater exactly.
tan(θ)
dy
| dx | of tan(θ) dy
| dx | of tan(θ) Back to the ideas of HOG, it decomposes gradient of a pixel
43 3 I(x, y) into a combination of two bins as described in Figure 1
244
tan(10) 17
and Equation 17. Hence, in each direction, the sum of two bins
56 97
57
tan(30) 168 equals to the gradient of the pixel. Equation 18 and 19 present
230
tan(50) 87 those equalities in the Ox and Oy axises, respectively. Finally,
193 73
250 11
the two magnitudes are the solutions of the two equations, and
tan(70)
91 4 have the form as Equation 20 and Equation 21, respectively.

Equation 14 leads to equivalent Equations 15 - 16 for deter- −−−−−→ −→ −−−→

mining two quantized orientations: θi and θi+1 , respectively. M (x, y)= Bθi + Bθi+1 (17)
It is clear from the two equations, there are only multiplication −−−−−→
||M (x, y)|| × cos α= dx
and comparison operations, and it is much simpler than inverse −→ −−−→
= ||Bθi || cos θi + ||Bθi+1 || cos θi+1(18)
tangent for hardware implementation. −−−−−→
||M (x, y)|| × sin α= dy
( d As −→ −−−→
| dxy | > = ||Bθi || sin θi + ||Bθi+1 || sin θi+1 (19)

Bs Bs × |dy | > As × |dx |
α> θ ⇔ d Ag ⇔ (15)
| dxy | ≥ Bg × |dy | ≥ Ag × |dx | −→ sin(θi+1 )dx − cos(θi+1 )dy
Bg ||Bθi ||= (20)
( d As sin(20)
| dxy | ≤

Bs Bs × |dy | ≤ As × |dx |
α< θ ⇔ d Ag ⇔ (16) −−−→ cos(θi )dy − sin(θi )dx
| dxy | < Bg
Bg × |dy | < Ag × |dx | ||Bθi+1 ||= (21)
sin(20)
where The advantages of this method is that it allows computing
A dy
• Bs
s
is the nearest smaller dx of tan(θ) value; gradient magnitude with only dx , dy information and the

978-1-5386-3210-9/17/$31.00 ©2017 IEEE 203

 2017 4th NAFOSTED Conference on Information and Computer Science

two quantized angles. The most complex part are sin, cos
values, but they are pre-computed. The accurate level of HOG
ca be controlled by choosing these values. To achieve these
advantages, this method requires determining the quantized
orientations exactly. If there is any errors in determining
direction of two bins, it produces negative magnitudes.
C. Proposed cell histogram generation scheme
Figure 2 shows the data flow of proposed cell histogram
generation, which uses our method. As shown in Figure
2, the input contains 4 neighbor pixels I(x + 1)(y), I(x −
1, y), I(x, y + 1), I(x, y − 1) of pixel I(x, y). In the first step,
the gradient in x and y directions dx and dy are computed. In
the second step, we have absolute values of dx and dy , and
the quadrant position of the gradient of pixel I. If the sign
−−−−−→
of dx is opposite to the sign of dy , the gradient M (x, y) is
in the second quadrant. Otherwise, it is in the first quadrant.
Because the left side of Oy axis is a reflection of the right side
of Oy, so computing the magnitude of two voted bins in the
first quadrant is enough. The third step is to determine the two
quantized angles θi , θi+1 via comparison between A × dx and
B × dy , where A is the dividend, and B is the divisor in Table
d
I. In this scheme, we use the nearest greater values dxy of tan θ.
I. After the orientations of two bins are determined, the two
−→ −−−→
corresponding magnitudes ||Bθi ||, ||Bθi+1 || are obtained from
Equations 20 and 21. Finally, we have to correct the angles
from quadrant position information. If those angles are in the
second quadrant, converting the θ in the first quadrant into the
second quadrant is done by a subtraction θ = 180 − θ.
As shown in Figure 2, we have already known one of Fig. 2. Dataflow diagram of the proposed method
two numbers in multiplication operations. It allows converting
all multiplications into form of shifts and additions. Table II TABLE II
C ONVERTING ALL MULTIPLY TO SHIFT AND ADDITION OPERATIONS WITH
shows all of the transformations with 8-bit length of fractional 8- BIT OF FRACTIONAL PART
part in sine and cosine. It is equivalent to multiply by 256. The
size of fractional part of magnitude is controlled by the sine, Shift and Shift and
B × dy A × dx
cosine values in the Table II. For example, if an application Addition Addition
needs a 10-bit length of fractional part, the cosine, sine values 4 22 11 23 + 21 + 20
have to multiply 1024 instead of 256. From Table II and
73 26 + 23 + 20 87 26 + 24 + 23 − 20
Figure 2, the total number of operations to identify two feature
168 27 + 25 + 23 97 26 + 25 + 20
bins of a pixel are about 30 additions and nearly 40 shifts.
17 24 + 20 3 21 + 20
IV. H ARDWARE IMPLEMENTATION AND EXPERIMENTAL cos(10) × 256 28 − 22 sin(10) × 256 25 + 23 + 2 2
RESULTS
cos(30) × 256 28 − 25 − 21 sin(30) × 256 27
We implemented the proposed method into hardware and cos(50) × 256 27 + 25 + 22 + 20 sin(50) × 256 27 + 26 + 2 2
examined its results. Figure 3 shows the hardware module
cos(70) × 256 26 + 24 + 23 sin(70) × 256 28 − 24 + 20
of the proposed method. It includes 4 input pixels I(x +
1, y), I(x−1, y), I(x, y+1), I(x, y−1) with 8-bit pixel length. cos(90) × 256 0 sin(90) × 256 28
1 15
The output contains the magnitude and the direction of two sin(20)
2+ 16
bins. Angle is 4-bit length data, and magnitude is 17-bit length
data with 8-bit of fractional part.
As shown in Figure 3, the module includes 4 steps. Firstly, The fourth step produces two voted bins, including orientation
the difference dx and dy are computed from the 4 input and magnitude.
pixels. In the second step the absolute values of dx , dy are Figure 4 presents the detail of the fourth steps. It has two
computed, and the quadrant position of the gradient of pixel multiplexers. The first multiplexer has 4 inputs from 4 angle
I(x, y) is detected. After having those absolute values, the pre- comparisons. This multiplexer chooses the highest angle θ in
calculating step calculates all the multiplications in Table II. the first quadrant. For each quantized angle, it also produces

978-1-5386-3210-9/17/$31.00 ©2017 IEEE 204

 2017 4th NAFOSTED Conference on Information and Computer Science

clk 𝑑𝑥
4 𝜃1
rst / 𝑑𝑦 𝑑𝑥′ − 𝑑𝑥
4
𝑒𝑥 =
𝜃2 𝑑𝑥
𝐼(𝑥 + 1, 𝑦) 𝑑𝑥 , 𝑑𝑦 , Angle and /
𝑑′𝑦 = 𝐵1 × 𝑠𝑖𝑛𝜃1 + 𝐵2 × 𝑠𝑖𝑛𝜃2
𝐵 × |𝑑𝑦 |
𝑑𝑥 , 𝑑𝑦 𝑞𝑢𝑎𝑑𝑟𝑎𝑛𝑡 𝐼𝐼 Magnitude 𝑑𝑦′ − 𝑑𝑦
𝐴 × |𝑑𝑥 |
𝐼(𝑥 − 1, 𝑦) Calculation 17 𝑑′𝑥 = 𝐵1 × 𝑐𝑜𝑠𝜃1 + 𝐵2 × 𝑐𝑜𝑠𝜃2 𝑒𝑦 =
𝐵1 𝑑𝑦
/
𝐼(𝑥, 𝑦 + 1) 17 𝐵2
/
𝐼(𝑥, 𝑦 − 1) Step 1 Step 2 Step 3 Step 4 Fig. 5. Error of calculation model

Fig. 3. Hardware implementation of the proposed method

V. C ONCLUSION

corresponding gradient magnitude from Equation 20 and 21. In this paper, we have proposed a new method to do
The second multiplexer converts the θ in first quadrant into cell histogram generation part without calculating the actual
the second quadrant if the sign(dx ) is opposite sign(dy ). gradient of a pixel. It identifies the two bins by the prop-
erties of decomposing gradient of a pixel into two standard
1 vectors. This method extracts HOG features of a pixel with
𝑞𝑢𝑎𝑑𝑟𝑎𝑛𝑡 𝐼𝐼 / 30 additions and 40 shifts. The error of the reconstructed
values of dx and dy from the two voted bins are always less
1 4 𝜃
4 1 than 2%. Additionally, it is very flexible, we only need to
17 × 𝑑𝑦 ≥ 3 × 𝑑𝑥 / /
/ define new constants for sine and cosine values of quantized
1 4 4
𝜃 angles to change the precision. The hardware implementation
168 × 𝑑𝑦 ≥ 97 × 𝑑𝑥 / / / 2
of the proposed method occupies about 3.57 KGEs with 45nm
1 17
𝐵1 NanGate standard cell library, at the maximum frequency
73 × 𝑑𝑦 ≥ 87 × 𝑑𝑥 / / 17
of 400MHz, and the throughput of 0.4 pixel/ns for a single
1 /
𝐵2 module.
4 × 𝑑𝑦 ≥ 11 × 𝑑𝑥 /

Fig. 4. Submodule angle and magnitude calculation
Table III shows the hardware implementation results of
some previous works and our proposed method. Pei-Yin and
Hsiao’s works calculated the gradient of current pixels with
approximate model via the SRA approximation for magnitude
and CORDIC for angle. Munteanu combined Equation 10 and
bin’s boundary angles method. As shown in the Table III, the
proposed method and Munteanu’s works can run at the highest
frequency of 400Mhz. They are also the two smallest areas cost
of implementation. However, the area of Munteanu’s module
is still more than five of our proposed method.
TABLE III
H ARDWARE COMPARISON OF C ELL H ISTOGRAM G ENERATION
Publications Technology Areas (KGates)
Pei-Yin [9] 130nm 85.5
Hsiao [8] 90nm 29.1
Munteanu [10] N/A 20
Our Proposed 45nm NanGate 3.57
Figure 5 shows our error of calculation model for the
proposed method. From the voted bin, we calculate a model
of dx and dy by Equation 18 and 19, called d0x and d0y . With
the d0x , d0y , dx , dy , the percentage of calculation errors at both
Ox and Oy directions is computed. Tested with all cases, the
results prove that our proposed method provides the percentage
of calculation errors ex ey , which are always less than 2% with
8-bit length of the fractional part.
R EFERENCES
[1] N. Dalal and B. Triggs, “Histograms of Oriented Gradients for Human
Detection,” in IEEE Computer Society Conference on Computer Vision
and Pattern Recognition (CVPR’05), vol. 1, pp. 886–893. [Online].
Available: http://ieeexplore.ieee.org/document/1467360/
[2] K. Negi, K. Dohi et al., “Deep pipelined one-chip FPGA implementation
of a real-time image-based human detection algorithm,” in 2011
International Conference on Field-Programmable Technology, pp. 1–8.
[Online]. Available: http://ieeexplore.ieee.org/document/6132679/
[3] K. Takagi, K. Mizuno et al., “A sub-100-milliwatt dual-core
HOG accelerator VLSI for real-time multiple object detection,”
in 2013 IEEE International Conference on Acoustics, Speech
and Signal Processing, pp. 2533–2537. [Online]. Available: http:
//ieeexplore.ieee.org/document/6638112/
[4] K. Mizuno, Y. Terachi et al., “Architectural Study of
HOG Feature Extraction Processor for Real-Time Object
Detection,” in 2012 IEEE Workshop on Signal Processing
Systems. [Online]. Available: https://pdfs.semanticscholar.org/f2c3/
9ff8ee36aca20915f6ce7195317b4e1a34cd.pdf
Frequency [5] R. Kadota, H. Sugano et al., “Hardware Architecture for HOG
167Mhz Feature Extraction.” IEEE, pp. 1330–1333. [Online]. Available:
N/A http://ieeexplore.ieee.org/document/5337209/
400Mhz [6] F. N. Iandola, M. W. Moskewicz et al., “libHOG: Energy-Efficient
400Mhz Histogram of Oriented Gradient Computation,” in IEEE 18th
International Conference on Intelligent Transportation Systems, pp.
1248–1254. [Online]. Available: http://ieeexplore.ieee.org/document/
7313297/
[7] A. Suleiman and V. Sze, “An Energy-Efficient Hardware Implementation
of HOG-Based Object Detection at 1080HD 60 fps with Multi-
Scale Support,” vol. 84, no. 3, pp. 325–337. [Online]. Available:
http://link.springer.com/10.1007/s11265-015-1080-7
[8] S.-F. Hsiao, J.-M. Chan et al., “Hardware design of histograms of
oriented gradients based on local binary pattern and binarization,”
in 2016 IEEE Asia Pacific Conference on Circuits and Systems
(APCCAS). IEEE, pp. 433–435. [Online]. Available: http://ieeexplore.
ieee.org/document/7803995/

978-1-5386-3210-9/17/$31.00 ©2017 IEEE 205

 2017 4th NAFOSTED Conference on Information and Computer Science

[9] P.-Y. Chen, Chien-Chuan Huang et al., “An Efficient Hardware

Implementation of HOG Feature Extraction for Human Detection,”
vol. 15, no. 2, pp. 656–662. [Online]. Available: http://ieeexplore.ieee.
org/document/6648678/
[10] “A Method for producing a histogram of oriented
gradients,” Patent Patent No. WO2 016 083 002. [On-
line]. Available: https://patentscope.wipo.int/search/docservicepdf pct/
id00000033608412/PAMPH/WO2016083002.pdf
[11] M. Peker, H. Altun et al., “Hardware emulation of HOG and AMDF
based scale and rotation invariant robust shape detection,” in 2012
International Conference on Engineering and Technology (ICET), pp.
1–5. [Online]. Available: http://ieeexplore.ieee.org/document/6396145/
[12] S. Bauer, S. Kohler et al., “FPGA-GPU architecture for kernel SVM
pedestrian detection,” in 2010 IEEE Computer Society Conference on
Computer Vision and Pattern Recognition - Workshops. IEEE, pp. 61–
68. [Online]. Available: http://ieeexplore.ieee.org/document/5543772/
[13] T. P. Cao and G. Deng, “Real-Time Vision-Based Stop Sign
Detection System on FPGA,” in 2008 Digital Image Computing:
Techniques and Applications. IEEE, pp. 465–471. [Online]. Available:
http://ieeexplore.ieee.org/document/4700058/

978-1-5386-3210-9/17/$31.00 ©2017 IEEE 206