<p>Urban pedestrian areas create hazardous conditions through their multiple street vendor and vehicular traffic and street infrastructure and pedestrian elements. The research presents a real-time object detection system which improves footpath surveillance and navigation support. The system implements a lightweight YOLOv8n (Nano) model which uses class refinement and audio signature integration to achieve better detection accuracy and enhanced user experience. The dataset contains footpath images which have been annotated to display multiple object categories which include pedestrians and vendors and vehicles and infrastructure and obstacles and trees. The researchers applied preprocessing techniques which included image resizing and noise reduction and pixel normalization and data augmentation to enhance their feature extraction process. The system uses convolutional layers to extract features at multiple scales while its optimized detection pipeline enables real-time object detection. The experimental results show that the system demonstrates better performance than the baseline systems. The proposed method achieves mAP@50 of 0.954, precision of 0.928, recall of 0.912, and F1-score of 0.920, while maintaining an inference latency of 8.4 ms and 119 FPS on CPU, demonstrating potential for real-time applications on CPU-based systems. The study showed that detection performance of MobileNet-SSD and YOLOv5n was inferior to the system because it achieved better detection results and faster processing times. The study introduces a new approach which combines lightweight object detection with audio feedback to create better human-computer interaction while delivering real-time guidance to help pedestrians navigate safely.</p>