Vehicle Detection Using YOLOv8 on Low-Resolution Images

Abstrak

Vehicle detection in low-resolution images remains a significant challenge in computer vision, particularly for embedded devices such as ESP32-CAM with limited computational resources and simple image resolution. This study evaluates the performance of YOLOv8 on low-resolution QVGA (320 × 240 pixels) images for vehicle detection and classification. The dataset was independently collected in a controlled laboratory environment using miniature vehicles, covering four vehicle classes (motorcycle, car, bus, and truck) with a total of 4,000 images and a 70:20:10 data split. A pretrained YOLOv8 model was fine tuned for 100 epochs and tested on an ESP32-CAM prototype. The evaluation results demonstrate excellent performance, achieving precision of 0.999, recall of 1.000, mAP@0.5 of 0.995, and mAP@0.5-0.95 of 0.995 on the validation data, as well as real-time detection accuracy of 97% for motorcycles and cars, and 99% for buses and trucks. These findings indicate that YOLOv8 can deliver reliable vehicle detection performance on low-resolution images and is suitable for implementation in embedded device-based systems