Pedestrian Detection

using FPGA

SAINATH REDDY C - 16ECE1006

MANIDEEP P - 16ECE1017
UNDER THE SUPERVISION OF
DR. VASANTHA MH
ECE DEPARTMENT
NIT GOA
Problem Statement

 The presence of cameras in our surroundings grows every day, allowing

object detection, and more specifically pedestrian detection, to be used in
a broad amount of applications

 A pedestrian recognition and warning system could save lives by warning

the driver about possible pedestrian hazard

 Due to their inherently parallel architecture Field Programmable Gate

Arrays(FPGA) are ideal for accelerating image processing tasks.
Algorithm

 Read an image from the camera

 Compute the Histogram of Oriented Gradients for the image

 Calculate the Feature Vector using Histogram of Oriented Gradients

 Pass the descriptor through Support Vector Machine(SVM)

 SVM is a classification machine learning algorithm

 SVM classifies whether there is human or not

Block Diagram

Host machine  DDR  Processing unit 

DDRRAM
 Processing unit:

Horizontal and Histogram Normali Feature SVM

Image  Resize(1:2) 
vertical gradients of 
gradients zation
 Vector
 
 Yes
data /NO
from
DDRRAM
How to calculate Histogram of
Oriented Gradients ?
Preprocessing:
 HOG feature descriptor used for pedestrian detection is calculated on a
64×128 patch of an image.
 Of course, an image may be of any size. Typically patches at multiple
scales are analyzed at many image locations.
 The only constraint is that the patches being analyzed have a fixed aspect
ratio.
 The patches need to have an aspect ratio of 1:2. For example, they can be
100×200, 128×256, or 1000×2000 but not 101×205.
 Calculate the Gradient Images
 To calculate a HOG descriptor, we need to first calculate the horizontal and
vertical gradients; after all, we want to calculate the histogram of gradients
 This is easily achieved by filtering the image with the following kernels

 For horizontal gradient image the kernel is

-1 0 1
 For vertical gradient image the kernel is

-1
0
1
Horizontal and vertical gradient images
of an example
Calculate Histogram of Gradients in
8×8 cells
 In this step, the image is divided into 8×8 cells and a histogram of
gradients is calculated for each 8×8 cells
 The histogram is essentially a vector ( or an array ) of 9 bins ( numbers )
corresponding to angles 0, 20, 40, 60 … 160.
Final Feature Vector

 To calculate the final feature vector for the entire image patch, the 9x1
vectors are concatenated into one giant vector.

 There are 15 horizontal and 31 vertical positions making a total of 15 x

31= 465 positions.

 Each 8x8 block is represented by a 9×1 vector. So when we concatenate

them all into one giant vector we obtain a 9x465= 4185 dimensional
vector.
Support Vector machine(SVM )
Support Vector Machine(SVM)

 Support Vector Machine (SVM) is a supervised machine learning algorithm which can

be used for both classification or regression challenges

 In SVM, we plot each data item as a point in n-dimensional space (where n is number

of features you have) with the value of each feature being the value of a particular
coordinate

 Then, we perform classification by finding the hyper-plane that differentiate the two

classes very well (look at the below snapshot)

 Support Vectors are simply the co-ordinates of individual observation. Support Vector
Machine is a frontier which best segregates the two classes (hyper-plane/ line).
ZYNQ Architecture
Work Done
Image after applying
Original Image Horizonatal edge kernel
Image after applying
vertical kernel Magnitude gradient
Direction gradient Final feature vector size
Future Work
 Validation of SVM and HOG descriptor using python present – Jan

 Start porting to FPGA Jan – Feb

1) Interfaces 2) HOG

 Interfacing Host machine and DDR Feb

HOG algorithm on FPGA

 SVM March
Interface of DMA from DDR to module

 Testing, Timing closure, data verification March – April

References:
THANK YOU