OPTICAL FLOW APPLIED TO TIME-LAPSE IMAGE SERIES TO ESTIMATE GLACIER MOTION IN THE SOUTHERN PATAGONIA ICE FIELD

ESTEBAN LANNUTTI; LUIS LENZANO; TOTH CHARLES; LENZANO MARIA GABRIELA

Prague

Congreso; XIII Congress. ISPRS. ISPRS; 2016

In this work, we assessed the feasibility of using optical flow to obtain data for estimating the movement of a glacier. In general, the former investigations used to detect glacier changes/motion involve solutions where the data set has an irregular point distribution and the cost to obtain those are high plus ground data tasks are necessary to support those approaches. Taking into account that glaciers are usually located in geographically complex and hard to access areas with challenging weather conditions, optical flow may provide an efficient solution at pixel level in temporal description of motion. A few studies in computer vision and image processing community used this method to detect large displacements, such as occuring with glaciers. Therefore, we carried out a test of the proposed large displacement optical flow (LDOF) method in Viedma Glacier, located at the Parque Nacional Los Glaciares, South Patagonia Icefield, Argentina. We collected a monoscopic image terrestrial sequence, time-lapse, acquired by a calibrated camera. Images were acquired every 24 hour from April 2014 until April 2015, a total of 362 days. The radiometric conditions change all the time due to various environmental factors, and thus a filter was applied to the images to minimize errors related to changes in lighting, shadows, clouds and snow. The filter is based on temporal correlation and RGB color discretization model between the images, allowing to select a sequence of correlated image pairs. This selection allowed discarding images that do not follow a sequence of similarity. Our results show an average range movement between 20-60 pixels per day, and the vector flow in the direction of the movement. In addition, we scaled the GSD by measurements taken on the ground. Finally, we analyzed the errors between different images pairs, and the matches generally appear to be adequate, although some areas show random gross errors related to the presence of changes in lighting. These errors were minimized by averaging the image sequence based on seasons, which yielded good results. The proposed technique allowed the determination of glacier motion during one year with accurate time-lapse interval, and produced accurate and reliable motion data for subsequent analysis.