Great advances have been made in the field of computer vision especially for object detection in the last two decades. Vehicle detection is fundamentally a task of object detection which assists in traffic control and management via traffic flow analysis. Models trained for vehicle detection on popular datasets containing vehicles as classes perform poorly while detecting vehicles in South Asian countries since the make and model of vehicles in the geographical region is different from those present in most popular datasets. Furthermore, some kinds of vehicles are found only in South Asian countries. In this paper, we propose a method that uses two YOLO models for the detection of various kinds of vehicles found on roads in South Asian countries. The first YOLO model detects vehicles while the second YOLO model detects wheels in case a truck is detected to further classify a truck based on the number of axles. A dataset consisting of images of bicycles, motorcycles, cars, buses, trucks, tempos and rickshaws was collected from surveillance videos, web scraping and subsetting the Indian Driving Dataset (IDD) as well as the Common Objects in Context (COCO) dataset. Another dataset was made which consists of images of trucks with their wheels annotated. YOLOv4 models trained on the latter mentioned datasets performed better than YOLOv3 and achieved a mAP of 71.08 per cent and 94.33 per cent for detection of vehicles and wheels, respectively. The proposed method to determine the number of axles of a truck gave an overall accuracy of 95 per cent.
Real-Time Detection of Vehicles on South Asian Roads
Lect. Notes Electrical Eng.
Machine Learning, Image Processing, Network Security and Data Sciences ; Kapitel : 31 ; 413-427
2023-01-01
15 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Engineering Index Backfile | 1910
|Planning Roads through Sensitive Asian Landscapes
Wiley | 2015
|Asian Highways and Rural Roads and Development
British Library Conference Proceedings | 1999
|Real-time Detection of Moving Vehicles
British Library Conference Proceedings | 1999
|Heavy vehicles and light roads
Engineering Index Backfile | 1933
|