Date of Publication
12-10-2022
Document Type
Bachelor's Thesis
Degree Name
Bachelor of Science in Electronics and Communications Engineering
Subject Categories
Electrical and Computer Engineering
College
Gokongwei College of Engineering
Department/Unit
Electronics And Communications Engg
Thesis Advisor
Leonard U. Ambata
Defense Panel Chair
Carlo Noel E. Ochotorena
Defense Panel Member
Roderick Y. Yap
Gerino P. Mappatao
Abstract/Summary
With the exponential rise in vehicular traffic volume, an intelligent system that is able to detect and classify would be essential. Image processing has already placed its significance for real-world applications on machine learning, one of which is traffic analysis. Through this study, the researchers developed a system that allows vehicular tracking and identification using the methods of neural networks for object detection, specifically the YOLOv5 algorithm. The process involved system training and testing programmed in the Python environment. The system training utilized an initial dataset generated by the researchers that eliminated the manual intervention of image annotation. The developed system was intervened under the conditions of two different locations. It showed its detection and recognition capabilities through the integrated system features, namely, counting, vehicle type classification, and traffic condition assessment. A region of interest (ROI) line was the primary basis for counting, and a lane-per-lane evaluation was displayed. These features were analyzed individually, between the actual and detected and wrong and correct detections. On the features of vehicle class and type identification, and detection count, the Video 1, Video 2 and Video 3 got an accuracy percentage of 83%, 90% and 90%, respectively, for vehicle classification, 94%,100%, and 97%, respectively for type of vehicle, 81%, 82% and 82%, respectively for color and 93%, 93% and 97% for detection count, respectively. The increasing trend on accuracy proves the efficiency of the fine-tuning process implemented on the system implementation except for the color that requires a separate training process. After the last finetuning process, an accuracy of 92.57% for vehicle class, 93% for type identification, 82% for
Abstract Format
html
Language
English
Format
Electronic
Keywords
Vehicle detectors; Image converters
Recommended Citation
Alfonso, L. S., Benitez, T. F., Cabalquinto, C. O., Perez, F. T., & Yang, J. (2022). Universal robust vehicle identification system. Retrieved from https://animorepository.dlsu.edu.ph/etdb_ece/20
Upload Full Text
wf_yes
Embargo Period
12-10-2022
