YOLOv8-based Pedestrian Detection for Malaysian Urban Traffic Surveillance: A Case Study at Bukit Bintang MRT

Abstract

The importance of pedestrian detection for intelligent transportation systems and city surveillance lies in the need for precise pedestrian identification to ensure road safety and optimize traffic efficiency. In this paper, deep learning techniques are investigated for detecting pedestrians in real-time video streams from urban traffic. The aim is to develop a pedestrian detection model based on YOLOv8, which can automatically detect pedestrians in surveillance videos recorded at the Bukit Bintang MRT crossroads. The detection model was trained and tested using 2,528 images captured and analyzed from traffic video streams. OpenCV and Ultralytics YOLOv8 framework were employed for data preprocessing, model training, and real-time inference. The experimental results prove that the YOLOv8 model has a precision of 78.40%, a recall of 74.50%, and an F1 score of 76.40%, with an AP value of 80.70% and a mAP value of 62.90%, respectively, under the same experimental conditions, which is better than the baseline of the YOLOv5s model. The proposed solution demonstrates the potential of deep learning models in the efficient control of pedestrian traffic, especially for safety applications in smart cities.