Motorcycles are the primary mode of transport in developing Asian countries; however, they are being accused of traffic congestion and accidents in urban areas, especially at signalized intersections. This study examined the recognition and identification of motorcycles on signalized intersection approaches. The study applied YOLOV8 instance segmentation model-a set of contours outlining each vehicle object in the image, a deep learning model for detecting and tracking vehicles, along with the deepsort (Simple Online Realtime Tracking) algorithm-to record the speed and movement path of motorcycles under mixed-traffic conditions in Chiang Mai city, Thailand. The study area presents a diverse array of two-wheeled vehicles. This has led to modifications and increased focus on learning about various types of two-wheeled vehicles model detection, such as motorcycles with passengers, motorcycles with non-helmet riders, delivery motorcycles, motorcycle trailers, and three-wheeled motorcycles (tuk-tuk). The performance of tracking algorithm was well depicted in the results, obtaining mAP and MAPE scores of models by 7.2% and 10.6% respectively for deepsort using data obtained from twelve signalized intersections. Additionally, the results also showed that the significant impact of motorcycle behavior and movement tracking on the traffic flows of intersection approaches. The tracking system's performance exhibited notable enhancements, with a 20.81% increase in Multiple Object Tracking Accuracy (MOTA) and a 7.82% increase in Multiple Object Tracking Precision (MOTP) respectively.
Using Instance Segmentation Model for Multi-Object Tracking of Motorcycles on Signalized Intersection Approaches
28.10.2023
1132200 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Storage Requirements for Signalized Intersection Approaches
Online Contents | 1996
|Examining Exposure of Motorcycles at Signalized Intersections
Transportation Research Record | 2008
|Maneuvers of motorcycles in queues at signalized intersections
Online Contents | 2012
|Right-Angle Crash Vulnerability of Motorcycles at Signalized Intersections
Transportation Research Record | 2010
|