A Study on Water Surface Floating Object Detection Based on YOLOV5s and Stereo Vision

Since the 1980s, the development of plastic products and economic growth have led to a significant accumulation of waste in rivers. Garbage disposal is urgent. This paper is based on the YOLOv5s object detection model as a preprocessing step, using a dataset of waterborne debris images collected from the internet. Common floating objects such as water bottles, milk cartons, branches, and white plastic bags were annotated using Labellmg(a tool for object detection annotation). The dataset was divided into training and testing sets, and multiple rounds of model training were conducted using PyTorch to select the best-performing model for deployment on mobile devices. Given the complexity of the water environment, including uncertain lighting conditions and numerous interferences, stereo vision is employed to ensure distance accuracy. By simulating human binocular vision, the system enables stereoscopic imaging, allowing accurate measurement of the distance between the camera and the center point of the object. Relying on the above algorithm and binocular camera, we can obtain the input image containing the target object, and obtain the type and distance of the debris on the water to provide convenience for the subsequent salvage work.