High Accuracy Estimation Of Head Yaw Using Bounding Box Algorithm On A Vision Based System

Tarmizi, Aine Ilina (2020) High Accuracy Estimation Of Head Yaw Using Bounding Box Algorithm On A Vision Based System. Masters thesis, Universiti Teknikal Malaysia Melaka.

	Text (Full Text) High Accuracy Estimation Of Head Yaw Using Bounding Box Algorithm On A Vision Based System.pdf - Submitted Version Restricted to Registered users only Download (2MB)
	Text (24 Pages) High Accuracy Estimation Of Head Yaw Using Bounding Box Algorithm On A Vision Based System.pdf - Submitted Version Download (670kB)

Abstract

Interaction between human and machines can be seen in various application nowadays such as robotic restaurant waiters, robots that assist elderly people and much more. The key to the success of these applications is the artificial intelligence of the machine that can comprehend the gestures of the human. In a face-to-face interaction, most applications use vision sensors such as camera as an input to the system. For the head movements of a human, shaking sideways is one of the common gestures used, which involve changes in head yaw angle. From the literatures, most researchers investigate on face recognition instead of head yaw or pose estimation. This research focuses on the head yaw estimation of a human from a faceto-face view of the camera. The purpose of this research is to establish an approach that can estimate the head yaw angle by using a single camera view. The objectives of the research are divided into three, to detect a face from the real-time stream of a camera and obtain bounding box coordinates around the detected face of the human, to analyze the change of distance and head yaw angle between the human face and camera with the effect of area and width of the bounding box on and lastly to design and validate the algorithm for estimation of the head yaw angle from the viewpoint of the camera. It was hypothesized that the width and area of the bounding box will decrease with the increase of head yaw angle. To achieve the objectives mentioned, the face detection was performed first, using Deep Learning method of Caffe framework which uses Convolutional Neural Network. By doing face detection, the bounding box was drawn around the face, which were used for the next analysis. From the video input of the camera, the changes in head yaw angle (orientation) incur changes in the bounding box location, width and area. The changes of the bounding box location, width and area were then analyzed to determine its relationship with the head yaw angle. The bounding box width and area were also analyzed with the different distance set between the face and the camera. In the head yaw estimation, the analysis performed earlier was used to estimate the head yaw angle by using the gradient and y-crossing values obtained. The hypothesis of the relation of the bounding box area and width with the head yaw angle was proven. The percentage of accuracy of the yaw estimation using a bounding box algorithm is 92.4%. In addition, the yaw estimation was also able to identify the turning direction of the head (left or right). A negative value of yaw angle shows the left turning direction while the positive value of yaw angle shows the right turning direction.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Image processing, Digital techniques, Neural networks (Computer science), Vision Based System
Subjects:	T Technology > T Technology (General) T Technology > TA Engineering (General). Civil engineering (General)
Divisions:	Library > Tesis > FKE
Depositing User:	F Haslinda Harun
Date Deposited:	18 Nov 2021 14:08
Last Modified:	18 Nov 2021 14:08
URI:	http://eprints.utem.edu.my/id/eprint/25402
Statistic Details:	View Download Statistic

Actions (login required)

View Item