286 0

Gender Classification of Passengers in a Vehicle Using Deep Learning

Title
Gender Classification of Passengers in a Vehicle Using Deep Learning
Other Titles
딥러닝을 이용한 차량내 탑승자의 성별 분류
Author
Minseok Kim
Alternative Author(s)
김민석
Advisor(s)
박태준
Issue Date
2019-02
Publisher
한양대학교
Degree
Master
Abstract
The development of the future car market is expected to be a smart car. In this regard, it is believed that the passenger in the car will be free from the driving task. Passengers can do another jobs likes watching movies, eating foods, having a meeting, etc. As a result, the space of automobile should be not a simple means of transportation. Therefore, the convenience service to provide passengers is essential part to the future car market. Especially if the services that match the characteristics of passengers likes gender, age, etc are more higher quality services. So, we propose the gender classification system of passengers in a vehicle. We apply hybrid networks to distinguish the gender of passengers in a vehicle in which upper body images of people entered in various face angles and vehicle environments. After that the face image is extracted from the upper body. To improve accuracy, the face data is augmented to determine the face image of various angles and environment based on the frontal image. Pre-trained CNN model extracts the augmented image data features more accurately. After that the extracted face features, we train the corelation of face data through artificial intelligence that grasps continuously changing characteristics called Long Short-Term Memory (LSTM). We conducted the experiment through Sorento vehicles data set and YouTube data set. As a result of real experiment, the performance result changes according to the sequence and combination of face data. The average accuracy is 88.3%, On the other hand, we have found the optimal data sequence and the optimal data combination. The optimal sequence is 5 and the combinations are 'original', 'rotate +30', 'rotate -30', 'Flip', and 'blurred'. We achieved a record of 94.5% .
URI
https://repository.hanyang.ac.kr/handle/20.500.11754/99519http://hanyang.dcollection.net/common/orgView/200000434543
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > INTERDISCIPLINARY ENGINEERING SYSTEMS(융합시스템학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE