ABSTRACT

Image Classification and detecting objects through machine learning is one of the
advancements in modern technology. Pedestrian Detection is also one of those extended
applications. However, there are many flaws in the existing systems of pedestrian
detection. An algorithim we proposed will benefit the flaws in existing systems. Where
we can detect the pedestrians using our system without a single flaw.

In our system we use convolutional neural network (CNN) for image classification
and R-CNN (Regional CNN) for object detection and IOU helps to measure the accuracy
of algorithms used for object detection using bounding boxes. Thus output is validated by
detecting the pedestrians exactly. This training cascade classifier improves the machine
to overcome the false detections near pedestrians. Experimental results on two widely
used pedestrian datasets demonstrate that the proposed training strategy and the CNN
based detector can effectively improve the detection rate and the localization accuracy
using fewer parameters.

v
 TABLE OF CONTENTS

CHAPTER NO. CHAPTER NAME PAGE NO.

ABSTRACT v
LIST OF FIGURES iv
1 INTRODUCTION

1.1 Aim of the Project 03

1.2 Outline of Project 03

1.3 Project Scope 04

1.4 Types Of Recommender Systems 04

1.5 System Features 07

2 LITERATURE SURVEY

2.1 Literature Survey 03

2.2 Existing System 04

2.3 Proposed System 05

2.4 Drawbacks Of Existing System 05

2.5 Advantages Of Proposed System 05

3 METHODOLOGY

3.1 Introduction To Python 06

3.1.1 What Can Python Do 06

3.1.2 Why Python 06

3.1.3 Good to Know 07

3.1.4 Python Syntax Compared

To Other Language 08

3.1.5 Python Is Interpreted 08

3.2 Regional CNN 09

R-CNN Architecture 10

Procedure For Object Detection 11

vi
 And Localization In R-CNN

Convolutional Neural Network 11

The Problem Space 12

Inputs And Outputs 13

Biological Connection 13

Structure 14

First Layer-Math Part 14

First Layer-High Level 16

Prespective

Disclaimer 18

Going Deepder Through The 18

Network

3.3.9 Fully Connected Layer 19

3.3.10 Training 21

3.3.11 Testing 23

3.3.12 How Companies Use CNNs 23

4 RESULTS AND DISCUSSION

Design And Implementation 27

Constraints

Constraints In Analysics 27

Constraints In Design 27

Constraints In Implementation 27

Other Non Functional Requirements 28

Performance Requirements 28

Safety Requirements 28

5 CONCLUSION AND FUTURE WORK

Conclusion 29

vii
Future Work 29

REFERENCES 30

APPENDIX 31

viii
 LIST OF FIGURES

FIGURE No. FIGURE NAME PAGE No.

3.1 R-CNN Architecture 10

3.2 CNN Classification 11
3.3 Internal Processing 12
3.4 Activation Map 15
3.5 Pixel To Visualization 16
3.6 Visuaization Of The Filter On Image 17
3.7 Summation Of Obtained Matrices 17
3.8 Working Of CNN 20
3.9 Recurrent Neural Network 24
3.10 Unrolled Recurrent Neural Network 24
3.11 System Architecture 25
4.1 Actual Output 26

ix
 CHAPTER 1
INTRODUCTION
Pedestrian detection has made significant advances benefiting from the region-
based convolutional neural networks. The project introduces a strict matching metric,
which is beneficial to selecting well-aligned positive samples A positive sample
matches a ground truth only if all its region-IoUs are bigger than a threshold. Secondly,
an improved negative example selection strategy using both the classification and
localization information is proposed to mine hard negative examples, which can further
suppress the false positive detections near the pedestrians. Recently, pedestrian
detection has made significant advances benefiting from the region-based
convolutional neural networks.
However, training R-CNN with a holistic intersection over union always brings
many flawed positive samples. This paper introduces a strict matching metric, which is
beneficial to selecting well-aligned positive samples. Specifically, this matching metric
is defined on a set of region-IoUs instead of a holistic IoU, which considers the
alignments of different part regions in a whole bounding box simultaneously.
A positive sample matches a ground truth only if all its region-IoUs are bigger
than a threshold. Secondly, an improved negative example selection strategy using
both the classification and localization information is proposed to mine hard negative
examples, which can further suppress the false positive detections near the
pedestrians. Based on the proposed sample selection strategy, a cascade compact
convolutional neural network is proposed for accurate pedestrian detection. Each stage
of the CC-CNN is constructed with a compact network that only consists of a small
number of parameters, thus making the detector suitable to be implemented on
onboard embedded systems.
To get well-aligned positive samples, this paper proposes a novel positive
example selection strategy based on the region wise metric, i.e., region-IoU which
considers the alignments of different part regions in a whole bounding box
simultaneously. First, this strategy defines a pedestrian template and a group of regions
based on the importance of pedestrian parts. Then a proposal bounding box is treated
as a positive sample if all its regions match the corresponding ones of any ground truth

1
with IoUs higher than a threshold. By this way, the well-aligned positive samples with
the existence of important parts are collected and used for training.
These unsatisfied performances are mainly due to two reasons. First, the
training methodologies, especially the positive example selection strategy, are
designed for general object detection, which are not appropriate for a specific category,
such as pedestrian detection. During the training process of single-stage/two-stages
CNN-based detectors most researchers select the positive examples whose IoUs
between anchors/proposals and ground truths are higher than a given threshold.
Traditional pedestrian detectors treat the ground truth bounding boxes as positive
examples during training. Different from them, the single-stage/two-stages CNN based
detectors select positive examples based on the overlap between anchor/proposal and
ground truth
To further suppress the false positives near the pedestrians, we also proposes
an improved negative example selection strategy which selects negative samples using
the fused score that combines both the classification score and the overlap value with
the ground truth.

THE MAIN CONTRIBUTIONS

(1) We are here to investigate the errors of the state-of-the art R-CNN based
pedestrian detector. The localization errors and the background errors are two main
failures for current R-CNN based pedestrian detectors. These false detections are
mainly caused by the flawed training samples extracted by the holistic IoU.
(2) A region-IoU based positive example selection strategy is introduced to get well-
aligned positive samples. Besides, the hard negative example mining method is
improved by combining both classification and localization information.
(3) A cascade compact network is proposed for pedestrian detection, in which each
stage is constructed by a compact network. It gains a high detection rate and
localization accuracy.
(4) The proposed detector achieves leading performances among the state-of-the-art
detectors on different data sets. Especially, it gains a significant improvement on
detection performance under the stricter localization criterion.

2
 CHAPTER 2
LITERATURE SURVEY
LITERARURE SURVEY

As a method to achieve real-time detection with high accuracy, cascade has been
widely used in pedestrian detection .They use a soft-cascade approach to transform a
classifier into cascaded ones to gain a detector with high detection performance and
efficiency Despite improvements in detection performance, it brings too many
parameters when multiple stages are included.

The proposed detector is similarly constructed in the form of cascade R-CNN

framework. Differently, benefiting from the modified example selection strategy, a
compact network is designed for each attached stage, which gains a similar detection
performance but saves the memory costs As a method to achieve real-time detection
with high accuracy, cascade has been widely used in pedestrian detection propose of
cascade face detection architecture built on CNNs operating at multiple resolutions. A
chained cascade network for object detection, where the early cascade stages are
used to reject obvious negative detections and the attached contextual stages are used
to learn more discriminative features for handling more difficult examples.

Both of them need extra proposal generation modules. propose of cascade R-CNN
architecture for high quality object detection. Despite improvements in detection
performance it brings too many parameters when multiple stages are included. The
proposed detector is similarly constructed in the form of cascade R-CNN framework.

As a method to achieve real-time detection with high accuracy, cascade has been
widely used in pedestrian detection They use a soft-cascade approach to transform a
classifier into cascaded ones to gain a detector with high detection performance and
efficiency. Most of the cascaded pedestrian detectors use hand-crafted features, such
as Haar-like features and HOG. Recently, the combination of the cascade framework
and CNNs has been studied in face detection and general object detection. The
following content has been studied in order to develop the accurate detection using
cnn,which is used here to develop the accurate pedestrian detection.

3
TITLE: STRENGTHENING THE EFFECTIVENESS OF PEDESTARIN DETECTION
WITH SAPITIAL POOLED FEATURES

AUTHORS: Sakrapee Paisitkriangkrai, Chunhua Shen, and Anton van den Hengel

YEAR: 2017

DESCRIPTION: We propose a simple yet effective approach to the problem of

pedestrian detection which outperforms the current state-of-the-art. Our new features
are built on the basis of low-level visual features and spatial pooling. Incorporating
spatial pooling improves the translational invariance and thus the robustness of the
detection process. We then directly optimise the partial area under the ROC curve
(pAUC) measure, which concentrates detection performance in the range of most
practical importance. The combination of these factors leads to a pedestrian detector
which outperforms all competitors on all of the standard benchmark datasets. We
advance state-of-the-art results by lowering the average miss rate from 13% to 11% on
the INRIA benchmark, 41% to 37% on the ETH benchmark, 51% to 42% on the TUD-
Brussels benchmark and 36% to 29% on the Caltech-USA benchmark.

TITLE: AN EMPIRICAL STUDY OF PEDESTRAIN DETECTION TECHNIQUES WITH

DIFFERENT IMAGE RESOLUTION

AUTHORS: Govardhan.S.D, Vasuki.A

YEAR: 2018 International Journal of Engineering and Advanced Technology (IJEAT)

DESCRIPTION: The designed technique identifies the pedestrian when the whole body
of pedestrians appears in system. Pedestrian detection is an essential issue in many
applications like robotics, video surveillance and computerized driver assistance.
Pedestrian detection techniques are enhanced one, but the detection results of current
systems are not adequate to be used in practice. The key issue is to identify the
pedestrians, despite large inconsistency of shape and appearance caused by
variations in pose, clothing, occlusion and lighting. The detection rate reduces when
vehicle or surrounding facilities cover person body parts. For addressing these issues,
many pedestrian detection techniques are employed for detecting part of an object.

4
 EXISTING SYSTEM

R-CNN analysis shows that localization error and background error are still the
two main errors for Faster R-CNN based detectors. In the following sections, the
proposed approach will deal with these two types of errors by introducing a new
example selection strategy and a cascade compact CNN. Besides, the annotations of
city person’s validation data set are manually re-annotated for fairer comparison of
different pedestrian detectors. Two types of errors are investigated in this section: false
positives and false negatives.In this system there are lost of problems regarding to
positive selection and negative selection, they are been mistakenly over lapping each
other to over come this method more efficient way has to be developed.

PROPOSED SYSTEM
A cascade compact convolutional neural network (CC-CNN) is proposed for
accurate pedestrian detection. CC-CNN based detector can effectively improve the
detection rate and the localization accuracy using fewer parameters. The experiment
is conducted on the City persons reasonable validation subset. All the false positives
are clustered into two categories: background errors and localization errors, To
simultaneously eliminate both localization and background errors, a new measurement
is introduced to select the negative samples. Specifically, the score of each negative.
DRAWBACKS OF EXISTING SYSTEM:
1.It provides plenty of distractions
2.It is an time effective method
3.Most of the pedestrains are not detected
4.Positives are also recognised as nagatives
ADVANTAGES OF PROPOSED SYSTEM:
1.Improved Image Classification
2.Improved Image Interpretation
3. Improved video classification Performance
4.Efficient at Delivering High-quality Results
5.Reduces the need for feature engineering
6. Strength to pure differences in the data is automatically learned

5
 CHAPTER 3
METHODOLOGY
INTRODUCTION TO PYTHON

Python is a widely used general-purpose, high level programming language. It

was initially designed by Guido van Rossum in 1991 and developed by Python Software
Foundation. It was mainly developed for emphasis on code readability, and its syntax
allows programmers to express concepts in fewer lines of code.

Python is a programming language that lets you work quickly and integrate systems
more efficiently.

It is used for:

 web development (server-side),

 software development,
 mathematics,
 System scripting.

WHAT CAN PYTHON DO?

 Python can be used on a server to create web applications.

 Python can be used alongside software to create workflows.
 Python can connect to database systems. It can also read and modify files.
 Python can be used to handle big data and perform complex mathematics.
 Python can be used for rapid prototyping, or for production-ready software
development.

WHY PYTHON?

 Python works on different platforms (Windows, Mac, Linux, Raspberry Pi, etc).
 Python has a simple syntax similar to the English language.
 Python is a language that we can use in integrated systems efficiently.