การตรวจจับและรู้จำตราสัญลักษณ์ของรถยนต์โดยใช้วิธีการโครงข่ายประสาทแบบคอนโวลูชันและพีระมิดฮิสโตแกรมของทิศทางเกรเดียนต์
Vehicle logo detection and recognition systems are developed for an automated vehicle information retrieval system that can be applied for tracking illegal activities such as searching for the stolen vehicles in the surveillance camera or traffic camera system. Recent studies show that most of logo...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Theses and Dissertations |
Published: |
เชียงใหม่ : บัณฑิตวิทยาลัย มหาวิทยาลัยเชียงใหม่
2020
|
Subjects: | |
Online Access: | http://cmuir.cmu.ac.th/jspui/handle/6653943832/69429 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Chiang Mai University |
Summary: | Vehicle logo detection and recognition systems are developed for an automated vehicle information retrieval system that can be applied for tracking illegal activities such as searching for the stolen vehicles in the surveillance camera or traffic camera system. Recent studies show that most of logo detection systems are using relative location between logo and license plate location. In this approach, the license plate has to be firstly detected. However, in some situations, license plates cannot be detected or the license plates are not located in the typical location. Although, many features have been proposed for vehicle logo detection and recognition, but the performances are still not good enough.
Therefore, the objective of this work is to improve the performance of the vehicle logo detection and recognition systems. This research involve the improvement of detection process by introduce the use of saliency map. The saliency map limits search area for logo localization which reduce the number of fault detection. In addition, a Convolutional Neural Network (CNN) is proposed in the feature extraction and classification step. Then, a Pyramid Histogram of Oriented Gradients (PHOG) is applied in the last step to verify the result from CNN. The experimental results show that the proposed process can achieve the better performance than local feature such as SIFT and using CNN or PHOG individually. |
---|