Evaluating the Performance of Ensembled YOLOv8 Variants in Smart Parking Applications Under Varying Lighting Conditions

With an emphasis on performance under various ambient illumination circumstances this paper explores the potency of YOLOv8 variants for vehicle and license plate detection. The suggested method will capture entire video frames, identify areas of interest with cars, and feed those regions into two distinct, pre-trained YOLOv8 models—one for license plate recognition and the other for vehicle detection. To make the photos easier for the Tesseract OCR engine to read, they are pre-processed using the OpenCV and Pillow libraries to make the images brighter and higher DPI. The four YOLOv8 models can be paired for vehicle and license plate identification tasks to produce sixteen possible combinations. We evaluate the performance of the chosen YOLOv8 combinations under various ambient light intensity levels (measured in lux) after they have been selected using TOPSIS analysis. Finding the most reliable model combinations that provide precise license plate and vehicle detection in the variety of illumination situations found in real-world settings is the goal of this evaluation.