Please use this identifier to cite or link to this item: http://prr.hec.gov.pk/jspui/handle/123456789/17148
Title: Computer Vision Based Automatic Traffic Sign Detection and Recognition for Developing World
Authors: , Abdul Mannan
Keywords: Engineering & Technology
Electrical & Computer Engineering
Issue Date: 2021
Publisher: University of Engineering & Technology, Lahore.
Abstract: For the last two decades, computer vision based techniques have got their way in sens ing road and road-side environment under the flag of intelligent transportation systems (ITS). One such application is automatic detection and recognition of road traffic signs for the purpose of assisting drivers and/or maintaining an inventory of road infrastruc ture. In the developing world, the problem becomes more challenging because of (1) the presence of inconsistent traffic signs due to weak implementation policies adopted for their fabrication (2) complex backgrounds rich in items having similar colors and shapes as that of the targeted traffic signs (3) occlusions/degradations caused by other nearby objects such as trees, poles, pedestrians, advertisement posters and aging etc. In order to solve the above three research problems, this work presents image processing, machine learning and deep learning based solutions. For the first research problem, a flexible spectral technique is applied ensuring maximum interclass separation and intraclass similarity. For research problems 2 and 3, a couple of approaches are presented for visible and partially occluded traffic signs. A completely data driven optimized custom color space solution is proposed to locate traffic signs in conjunction with energy compaction based spectral augmentation. To further enhance the effectiveness of the proposed technique, an automatic emphasis adjustment strategy is also introduced. Research problem 3 is also addressed with the help of a continuously updating flexible Gaussian mixture model with the ability to split and merge. The technique is inspired of human cognition and employs a subspace based convolutional spectral approach to classify traffic signs partially occluded or degraded. The method ensures higher execution speed coupled with enhanced accuracy on test data. The field of automatic traffic sign detection and recognition has been suffering with unavailability of benchmark datasets. A few datasets containing only the visible traffic signs are available for some European countries. As part of this project, datasets of visible, occluded and degraded traffic signs were collected/synthesized as a result of field surveys on N5 and M2 highways in Pakistan. A set of experiments performed on benchmark and self collected datasets reveal that the proposed strategies for the detection and recognition of inconsistent, occluded and degraded traffic signs outperform state of the art and recently published methods.
Gov't Doc #: 23362
URI: http://prr.hec.gov.pk/jspui/handle/123456789/17148
Appears in Collections:PhD Thesis of All Public / Private Sector Universities / DAIs.

Files in This Item:
File Description SizeFormat 
abdul mannan electrical engg 2021 uet lhr.pdfphd.Thesis8.93 MBAdobe PDFView/Open


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