Improved Vehicle Traffic Detection Using YOLO Algorithm with Brightness and Contrast Adjustment in Low-light Conditions

This study aims to improve YOLO (You Only Look Once) object detection in low light conditions by improving image quality using techniques such as CLAHE, brightness/contrast adjustment and noise reduction.CLAHE is expected to improve local contrast in the image, while brightness and contrast adjustment will be used to adjust the overall brightness and contrast. Noise reduction will be used to reduce distracting noise in the image. Experiments will be conducted using YOLOv5 and YOLOv8 to illustrate the effectiveness of the proposed approach.Mean Absolute Percentage Error (MAPE), F1 score and confusion matrix are used to measure the prediction accuracy. The results of the evaluation method using MAPE showed that the manipulated YOLOv5s produced a score of 0.59, which was higher than the raw YOLOv5s which produced a MAPE score of 0.63. The MAPE score for the manipulated YOLOv8s resulted in a score of 0.49 which was better than the MAPE score for the raw YOLOv8s which was 0.52, concluding that the lower the MAPE value, the better the accuracy of YOLO object detection. The F1 score for the enhanced YOLOv5s is 0.12, which is higher than the score for the pre-enhanced YOLOv5s which is 0.10, and for the YOLOv8s the F1 score is 0.19, which is higher than the non-enhanced model which is 0.18. The results suggest that adjusting image brightness, contrast and noise reduction improves target detection in low light conditions.