Robot–Enabled Lakeshore Monitoring Using Visual SLAM and SIFT Flow

This paper establishes an autonomous monitoring framework to augment a human’s ability to detect changes in lakeshore environments. This is a large spatial and temporal scale study, which analyzes data from eight different surveys of a lakeshore collected over 11 months with an autonomous surface vehicle. Despite the variation in appearance across surveys, our framework provides a human with aligned images and a way to readily detect changes between them. First visual SLAM is used to find a coarse alignment of images between surveys, and second, SIFT Flow is applied to achieve dense correspondence. The aligned images are flickered back-and-forth in a user display, which enables a human to rapidly detect changes. Results show our method can align images in the midst of variation in appearance of the sky, the water, changes in objects on a lakeshore, and the seasonal changes of plants.