Real-Time 3D Sonar Image Recognition for Underwater Vehicles

The recognition of objects and environment are one of keys to automate underwater tasks. This paper presents a predictor based image recognition method using high-resolution sonar images. They are very sensitive to sonar's view point changes. The proposed predictor predicts an object images depending on the view point. Using the object model, it modifies a template for optimal recognition. To estimate the accuracy and reliability of the proposed method, indoor tank tests were carried out using various shapes of objects.