Comparison Of SSD Method Performance with YOLOv7 in Detecting River And Lake Traffic Signs

Abstract

River and lake traffic signs play an important role in maintaining the safety of water navigation, but there is still a lack of understanding and supporting technology to recognize them automatically. This study aims to compare the performance of two popular object detection methods, namely YOLOv7 and Single Shot Multibox Detector (SSD), in identifying river and lake signs based on digital images. The dataset used consists of 6,650 images, including original data and data that has been augmented with visual degradation such as low resolution, noise, blur, lighting artifacts, occlusion, and cluttered background. This study evaluates the models based on accuracy, precision, and recall metrics. The test results show that YOLOv7 excels in all metrics, with the highest accuracy of 78%, compared to SSD which only reaches 70%. Both models experience a significant decrease in performance under occlusion and cluttered background conditions, but YOLOv7 still shows better resilience. Meanwhile, the blur condition is the degradation that least affects the performance of both models. These findings confirm that YOLOv7 is more robust to visual disturbances and can be a more effective alternative in automatic river and lake traffic sign recognition systems.