| dc.description.abstract | Dhaka City experiences persistent traffic congestion due to increasing vehicle density, limited road
capacity, and the complexity of its mixed-traffic environment. Auto-rickshaws are a common
component of urban transportation, making their automated detection valuable for real-time
traffic monitoring and intelligent transportation systems. This study presents a lightweight
YOLO-based urban surveillance framework for real-time auto-rickshaw detection in Dhaka City.
An Auto-RickshawImageBD dataset containing 1,331 original images was used as the primary data
source. Through data augmentation using the Albumentations library, the dataset was expanded to
3,671 images. The images were annotated into two classes, namely auto_rickshaw and
non_auto_rickshaw, and prepared in the YOLO object-detection format. Three lightweight YOLO
models, YOLO11n, YOLO12n, and YOLO26n, were trained and comparatively evaluated using
Precision, Recall, mAP@0.5, mAP@0.5:0.95, and F1-score. The experimental results show that
YOLO11n achieved the best overall detection performance. On the test dataset, YOLO11n
obtained an mAP@0.5 of 83.93%, mAP@0.5:0.95 of 47.64%, Precision of 84.19%, Recall of
73.82%, and F1-score of 78.66%. YOLO26n achieved the second-best performance, while
YOLO12n produced comparatively lower detection results under the evaluated configuration.
Based on these findings, YOLO11n was selected as the primary model for the proposed lightweight
surveillance framework. The framework is intended to process traffic-camera imagery and provide
vehicle detection information that can support traffic-density estimation and congestion alerts. The
study demonstrates that a lightweight YOLO model can provide an effective balance between
detection accuracy and computational efficiency for auto-rickshaw monitoring in the
challenging urban traffic environment of Dhaka City. | en_US |