You are on page 1of 17

Impact of Enhanced Cascade Using Histogram of Oriented

Gradients on Pedestrian Detection

A Research Presented to the Faculty of

College Department

Emilio Aguinaldo College-Cavite

In Partial Fulfillment of the

Requirements for Mechanical Engineering

Submitted By:

Agot, Ralph Brenz

Baula, Bryan Cris Matthew

Flores, James Nicolai

Maliksi, Yvan Eric

Orias, Chryss Adrian


CHAPTER 1
THE PROBLEM AND ITS BACKGROUND

INTRODUCTION

In recent years, pedestrian detection has been an active area of research due to

the potential positive impact of related applications on quality of life, including

surveillance systems, automotive safety, robotics, multimedia content analysis, assistive

technology, and advanced interactive interfaces. Detecting pedestrians is one of the

most difficult challenges in computer vision due to the wide variety of human

appearances, poses, and environmental conditions. Even though numerous significant

approaches have been proposed in recent years, pedestrian detection still offers a vast

enhancement framework, particularly in terms of accuracy and efficiency.

According to Fu, et.al, (2013), it is difficult to reliably detect individuals in a video

sequence due to numerous obstacles. Histograms of Oriented Gradients (HOG) is one

of the most effective discriminative characteristics for locating persons. Although the

HOG feature encodes the main contour information well, the background noise

degrades the gradient information. Moreover, detecting humans in images is a difficult

endeavor due to their varying appearances and a wide variety of poses. The first

requirement is a robust feature set that enables the human form to be discerned

precisely even in cluttered backgrounds with poor illumination. We investigate the issue

of human detection feature sets and demonstrate that locally normalized Histogram of

Oriented Gradient (HOG) descriptors outperform other existing feature sets, such as

wavelets. The proposed descriptors resemble edge orientation histograms, SIFT


descriptors, and shape contexts, but they are computed on a dense grid with uniformly

spaced cells and use overlapping local contrast normalizations for enhanced

performance.

Recently proposed by Dalal and Triggs, the locally normalized histogram of

oriented gradients (HOG) descriptor is one of the most widely used feature extraction

techniques for pedestrian detection. Using HOG to extract gradient features from

preprocessed Gabor images, the enhanced human detection method Hon bGG is

presented. Zhu et al. proposed a method for rapid human detection by combining a

cascade of rejectors with HOGs of variable-size blocks to extract distinctive features.

The extraction of pedestrian features is a crucial component of pedestrian

detection. In recent years, because of the development of deep learning technology,

researchers have begun to use deep neural networks, particularly convolutional neural

networks, to automatically extract features from original image inputs. Deep learning

methods can obtain higher-level semantic features than conventional, artificially

designed feature detection techniques. The method of pedestrian detection based on

artificially designed features first extracts pedestrian features based on a specific area

in the image and then transmits the extracted features to a previously trained detector to

determine whether a pedestrian is present. Feature extraction, classifier training, and

detection are performed independently. Pedestrian detection methods based on deep

learning can integrate feature extraction, classifier training, and detection processes into

a framework to accomplish an end-to-end approach from feature extraction to detection.


BACKGROUND OF THE STUDY

Computers have become indispensable in our daily lives; they are useful in many

areas because they can perform complex and specific tasks more efficiently and

accurately than humans. Nonetheless, they are not yet capable of performing intelligent

high-level tasks, such as scene analysis, speech recognition, reasoning, and logical

interpretation, that humans perform frequently and unconsciously.

Among the cutting-edge areas of current research is the creation of vision

systems with human cognitive capabilities. Object recognition is one of these capacities;

a human can classify an object with ease and speed at a single glimpse. In addition,

humans can recognize objects, such as cars and people, regardless of their hue, pose,

texture, deformation, occlusions, illumination, or background complexity. Numerous

researchers in the field of computer vision are concentrating on the creation of object

recognition systems able to determine the precise identity of an observed object.

Implementing accurate object detection, which entails locating a specific object in a

static image in terms of its position and size, is one of the primary tasks required to

achieve this objective.

Pedestrian detection is a specific type of object detection in which the objective

object class is people. It has recently been a popular area of study due to the

prevalence of applications involving the tracking of people's location and movement, as

well as the potential positive impact of applications derived from it, such as surveillance

systems, automotive safety, robotics, multimedia content analysis, assistive technology,

and advanced interactive interfaces. Despite the fact that numerous approaches have
already been proposed, pedestrian detection remains a difficult undertaking due to the

vast diversity of pedestrian appearance. In addition, it is a competitive domain because

the high research activity in the field continually pushes the upper limit of precision and

efficacy to new heights. Existing pedestrian detection methods combine fundamental

features with trainable classifiers.

STATEMENT OF THE PROBLEM

The purpose of the present thesis is to develop and evaluate a state-of-the-art

pedestrian detection approach. In particular, the work will pursue two different goals:

Building a pedestrian detector from scratch and evaluating its performance, both in

terms of accuracy and speed. In this study, the approach uses an enhanced cascade

using histograms of oriented gradients on pedestrian detection. It aims to study the

impact of various feature extraction, HOG, and classification techniques in arriving to a

pedestrian or non-pedestrian decision.

1. What is the impact of using histogram of oriented gradients on the development

of pedestrian detection?

2. Is there a significant relationship between the pedestrian detection data set used

in the pedestrian decision?

THEORETICAL FRAMEWORK

The LSVM (Linear Support Vector Machine) is primarily based on the SRM

(structural analysis risk minimization) principle. It maps the data to the high-dimensional

feature space via the kernel function mapping, finds the optimal linear hyperplane in the
high-dimensional feature space, separates the two categories of data, and maximizes

the classification interval between the two types. The Mercer condition of the kernel

transforms the optimization issue into a convex quadratic programming issue.

Dalal and Triggs demonstrated in their 2005 paper, Histogram of Oriented

Gradients for Human Detection, that the Histogram of Oriented Gradients (HOG) image

descriptor and a Linear Support Vector Machine (SVM) could be used to train highly

accurate object classifiers — or in their study, human detectors.

HYPOTHESIS

Pedestrian detection can be understood as a

combination of two building blocks, a feature extraction algorithm, and a classification

method that uses the features to make the object/non-object decision. With this, the

following null and alternative hypotheses were drawn:

1. There is no impact of using histogram of oriented gradients on the development

of pedestrian detection.

2. There is no significant relationship between the pedestrian detection data set

used in the pedestrian decision.


SCOPE AND LIMITATIONS

This study is limited only to the impact of enhanced cascade using histogram of

oriented gradients on pedestrian detection. This is to determine the performance of

feature extraction and classification techniques used in having a pedestrian or non-

pedestrian decision. The findings of this study will only apply to assess the

influence of different training parameter values on the performance of a pedestrian

detector. Thus, this study reveals the structure and performance of the obtained

pedestrian detector and proposes a sensitivity analysis, explaining all the evaluated

configurations and important experimental results.

RELATED LITERATURE AND STUDIES

In the study of Ahlberg et.al (2014) entitled “Pedestrian Detection using a

boosted cascade of Histogram of Oriented Gradients” from the study it aims to influence

the different training parameters values on the performance of a pedestrian detector.

First, a pedestrian detector, using a boosted cascade of Histograms of Oriented

Gradients, is built from scratch. Afterwards, a sensitivity analysis is carried out taking

into account significant variables, such as the number of training samples, the feature

(HOG) or classifier (SVM) parameters, the feature selection technique, the negatives

resampling treatment and the typology of the employed weak classifier. It was found out

that although many significant approaches have been proposed lately, pedestrian

detection still offers a wide framework of improvement, mainly in terms of accuracy and

efficiency.
According to the study of Wang et.al (2009) entitled “Pedestrian Detection

System Using Cascaded Boosting with Invariance of Oriented Gradients” the study aims

to present a novel learning-based pedestrian detection system capable of automatically

detecting individuals of different sizes and orientations against a wide variety of

backgrounds, including crowds, even when the individual is partially occluded. Wherein

the study found out, that the use of the Gaussian-weighted window approach and tri-

linear interpolation technique in constructing the HOG feature vectors improves the

detection performance from 91% to 94.5%. In the proposed scheme, the detection

process is performed using a cascaded detector structure in which the weak classifiers

and corresponding weights of each stage are established using the AdaBoost self-

learning algorithm. In addition, the results revealed that the cascaded structure not only

provides a better detection performance than many of the schemes presented in the

literature, but also achieves a significant reduction in the computational time required to

classify each input image.

In the study of Ragb & Asari (2016) entitled “Histogram of Oriented Phase and

Gradient (HOPG) Descriptor for Improved Pedestrian Detection” the study aims to

present a new pedestrian detection descriptor named Histogram of Oriented Phase and

Gradient (HOPG) based on a combination of the Histogram of Oriented Phase (HOP)

features and the Histogram of Oriented Gradient features (HOG). The proposed

descriptor extracts the image information using both the gradient and phase congruency

concepts. Wherein their experimental study result shows that the human detection

system based on the proposed features has less error rates and better detection

performance over a set of state-of-the-art feature extraction methodologies. According


to the study of Sancho (2014) entitled “Pedestrian Detection using a boosted cascade of

Histogram of Oriented Gradients” he depicted in his study he influences of different

training parameters values on the performance of a pedestrian detector. Wherein the

process includes first, a pedestrian detector, using a boosted cascade of Histograms of

Oriented Gradients, is built from scratch. Afterwards, a sensitivity analysis is carried out

taking into account significant variables, such as the number of training samples, the

feature (HOG) or classifier (SVM) parameters, the feature selection technique, the

negatives resampling treatment and the typology of the employed weak classifier.

In the study of Wei et.al (2013) entitled “An Improved Pedestrian Detection

Algorithm Integrating Haar-Like Features and HOG Descriptors” their study depicted on

that there is importance of pedestrian detection in a variety of applications such as

advanced robots and intelligent surveillance systems. Wherein from their study it

depicted on how they can improve pedestrian detection method through integrating

Haar-like features, AdaBoost algorithm, histogram of oriented gradients (HOG)

descriptor, and support vector machine (SVM) classifiers, in which the head and

shoulder information is utilized especially. Due to the fast training speed of Haar-like

features and the high detection efficiency of HOG features, the proposed method can

classify pedestrians precisely with higher speed. From their experimental results it

validated that the efficiency and effectiveness of the proposed algorithm.


Chapter 2

Research Methodology

The study will be using the method of experimental investigation, in order to

develop and evaluate a state-of-the-art pedestrian detection. Experimental studies are

ones where researchers introduce an intervention and study the effects. Experimental

studies are usually randomized, meaning the subjects are grouped by chance (Institute

for Work and Health, 2016). More so, experimental investigations involve a process in

which a "fair test" is designed and variables are actively manipulated, controlled, and

measured in an effort to gather evidence to support or refute a causal relationship

(Texas Education Agency, 2007).

Research Design

The study will be utilizing the Control group design to measure and evaluate the

impact of using histogram in developing pedestrian detection and to know the

relationship between the pedestrian detection data set uses in the pedestrian decision

within the two groups and to come up with decision regarding pedestrian detection.

Control groups are designed to measure research bias and measurement effects, such

as the Hawthorne Effect or the Placebo Effect. A control group is a group not receiving

the same manipulation as the experimental group. Experiments frequently have 2

conditions, but rarely more than 3 conditions at the same time (Blaskstad, 2020).
Sampling Techniques

The random sampling was used in the selection of the respondent who used to

answer the questionnaire. Random sampling is a randomly selected subset of a

population. In this sampling method, each member of the population has an exactly

equal chance of being selected.

This method is the most straightforward of all the probability sampling methods,

since it only involves a single random selection and requires little advance knowledge

about the population. Because it uses randomization, any research performed on this

sample should have high internal and external validity (Thomas, 2020).

Survey Instruments

A data collection instrument was used in the study, namely the survey

questionnaire. The objectives of the study as well as the research question have been

used to guide the formulation of items in the instruments. After the implementation of

the building of pedestrian detector, the researcher will now proceed to the evaluation

through answering the survey questionnaire that will be given to the respondent in order

to investigate and gather the needed data. The survey questionnaire is composed of 2

questions answered by the chosen respondent on the random sampling. Part I of the

survey composed of questionnaire about the impact of using histogram of oriented

gradients on the development of pedestrian detection. Part II is composed of the

relationship between the pedestrian detection data set used in the pedestrian decision.
Data Gathering Procedure

The researchers undertook the following steps and procedures in gathering data

and information needed for the study:

1. The researchers secured the approval of the Department Chairperson and

the Dean of the College of in conducting the study.

2. The researchers prepared the survey instrument upon the Department

Chairperson and College Dean's approval.

3. The researcher will now implement the building of pedestrian detector and

evaluate it’s use.

4. The researchers tested the validity and reliability of the instrument through a

pre-test.

5. After the validity and reliability of the instrument were ensured, the

researchers sent a letter to the Dean and the Department Chair seeking

permission to conduct data gathering for the study.

6. Upon approval of the request, the researchers utilized available electronic

media (i.e., Google Form) to send the instrument to the identified respondents

of the study. In the conduct of data gathering, the researchers ensured that

the following ethical guidelines would be observed:

a. Informed Participation and Voluntary Consent. The researchers

ensured that participants were aware of their participation in the study,

informed of the consequences, and consented to participate. No one also

will be forced to participate in the study.


b. Privacy, Confidentiality, and Safety. The researchers informed the

respondents that their answers and responses were being recorded and

assured them that their personal information and responses would be

confidential.

c. No harm to the participants. The researchers ensured that the data

analysis and reporting of the study would not harm the participants. The

researchers also used careful terms and phrases to avoid offending the

respondents.

7. After gathering the needed data, the researchers tallied, processed,

interpreted, and analyzed the data. This was where the results, conclusions

and recommendations were based from.

Measuring Scales to rate each statement

Scale Response Anchor Interpretation

5 Strongly Agree Very Highly

Implemented

4 Agree Highly Implemented

3 Neither Agree nor Disagree Implemented

2 Disagree Poorly Implemented

1 Strongly Disagree Not Implemented


Statistical Treatment of data

The information gathered were tabulated and processed manually with the aid of

the computer. Data analysis was done to produce the findings and result of the study.

All data were analyzed by summing-up at the responses and are presented in the table

format. The total counts for each rating were based using the Likert scale.

The weighted mean is a method for combining the means of two or more groups

of varying sizes, taking their sizes into account when computing the overall or grand

mean (Taylor, 2023).

Multiple Regressions a set of procedures for investigating the straight-line

relationships between two or more variables is referred to as analysis. It's a more

complicated version of simple linear regression. When we wish to anticipate the value of

a variable based on the values of two or more other variables, we utilize this method

(Hayes, 2023).

Pearson R Moment Correlation is a measure of the strength and direction of

association that exists between two variables measured on at least an interval scale. It

is the test statistics that measures the statistical relationship, or association, between

two continuous variables. It is known as the best method of measuring the association

between variables of interest because it is based on the method of covariance ( Laerd

Statistics, 2018).
REFERENCES

Ahlberg, J. & Markus, N. (2014). Pedestrian Detection using a boosted cascade of

Histogram of Oriented Gradients. Universuty Politecnica De Catalunya.

Wang, C.R. Wu, J. & Lien, J.J. (2009). PEDESTRIAN DETECTION SYSTEM USING

CASCADED BOOSTING WITH INVARIANCE OF ORIENTED GRADIENTS.

International Journal of Pattern Recognition and Artificial Intelligence, pp. 801-

823.

Ragb, H. & Kasari, H. (2016). Histogram of Oriented Phase and Gradient (HOPG)

Descriptor for Improved Pedestrian Detection. University of Dayton eCommons .

Tea. (2007). Scientific Reasoning - Planning Comparative and Experimental

Investigations. Retrieved from Texas Gateaway:

https://www.texasgateway.org/resource/scientific-reasoning-planning-

comparative-and-experimentalinvestigations#:~:text=Experimental

%20investigations%20involve%20a%20process,does%20not%20receive%20any

%20treatment.

Blaskstad, O. (2020). Experimental Research. Retrieved from Explorable:

https://explorable.com/experimental-research
Thomas, L. (2022, December 05). Simple Random Sampling | Definition, Steps &

Examples. Scribbr. Retrieved May 22, 2023, from

https://www.scribbr.com/methodology/simple-random-sampling/

Taylor, S. (2023, May 7). Weighted Mean. Retrieved from CFI:

https://corporatefinanceinstitute.com/resources/data-science/weighted-mean/

Hayes, A. (2023, April 29 ). Multiple Linear Regression (MLR) Definition, Formula, and

Example. Retrieved from Investopedia:

https://www.investopedia.com/terms/m/mlr.asp

Laerd Statistics. (2018). Retrieved from Pearson's Product-Moment Correlation using

SPSS Statistics: https://statistics.laerd.com/spss-tutorials/pearsons-product-

moment-correlation-using-spss-statistics.php
PROPOSED TITLE

1. Impact on Delay Period of Performance of Compression Engine Running on Jatropha


Fuel
2. The experimental investigation of the performance evaluation of three charcoal stoves
3. Evaluating the impact of usage on the fabrication electrical Pop Corn Machine

You might also like