Glacial Ice Detection - Stacks2012. References The k-means classification technique employed on...

Glacial Ice DetectionAshley Clark, Tyler Reid, & Paul Tarantino

Dept. of Aeronautics & Astronautics, Stanford University, Stanford CA 94305

Introduction Image DataThe repeatability plots below show the robustness of the clusteringscheme under different simulated conditions such as brightness, cameracontrast, and camera resolution. The false color images furtherdemonstrate robustness under harder to model conditions, such asillumination angles and perspective shifts.

Robustness

[1] T. Sundlisæter, T. Reid, C. Johnson, and S. Wan, "GNSS and SBAS System of Systems: Considerations for Applications in the Arctic," in 63rd InternationalAstronautical Congress, Naples, Italy, 2012.

[2] United States Geological Survery, "90 Billion Barrels of Oil and 1,670 Trillion Cubic Feet of Natural Gas Assessed in the Arctic," ed, 2008.[3] NSIDC. (2011). Summer 2011: Arctic sea ice near record lows. Available: http://nsidc.org/arcticseaicenews/2011/10/summer-2011-arctic-sea-ice-near-record-lows/[4] N. Kjerstad, Ice Navigation: Akademika Publishing, 2011.[5] Icebreaking Program Maritime Services Canadian Coast Guard Fisheries and Oceans Canada, "Ice Navigation in Canadian Waters," ed. Ottawa, ON, Canada,2012.

References

The k-means classification technique employed on images of theHelheim Glacier demonstrate robust detection of signs of glacial ice. Thisshows promise in the augmentation of marine radar-only systems with acomputer vision system for improved safety of operation.

Conclusion

The authors would like to greatly acknowledge Ananda Fowler of RIEGL LMSfor providing us with this data and his continued support on this project. Wewould also like to credit images LeWinter #’s 06, 07, and 82 to AdamLeWinter, USACE CRREL.

Acknowledgements

These images of the Helheim Glacier allow us to test the robustness of our glacialice detection scheme against different camera angles, lighting, and resolutions.

LeWinter #06 – Nominal image for comparison.

LeWinter #07 – 91.6% repeatability with image #06 #1253 – 63.9% repeatability with image #06Classification Results

Applying the threshold specified by the k-means clustering analysis on imageLeWinter #82 results in the false colour image below. Pink represents signs ofpotential danger: either patches of land or glacial ice not covered by snow.Images of the Helheim Glacier in Greenland were utilized for this

analysis (see images on the right).The area photographed consists ofbanks of land, glacial ice, fresh snow, and many other materials. Thoughit is mostly all ‘glacial ice’, it is mainly covered in snow. Thus what weare interested in is our ability to detect the reflectance signature ofglacial ice where it is not covered by snow and highlight it as a warningof its presence. Using the known

k-Means Clustering

LeWinter #82 LeWinter #06

LeWinter #07 #1253

Safe marine navigation in the Arctic is becoming more important with agrowing interest in the region in recent years [1]. The United StatesGeological Survey (USGS) estimates there to be more than 90 billionbarrels of undiscovered oil in the arctic [2]. With the summer Arctic seaice extent having decreased by 50% since 1980 [3], this now openingwaterway has given rise to serious interest in commercial exploitation ofits resources as well as shipping routes through the Arctic.

There are several challenges that face ships operating in Arctic waters,one of which is the constant danger of multi-year and glacial icecollisions [4]. Knowledge of its whereabouts is crucial to safeoperations. Radar is a useful tool but it is strongly advised not to relysolely on this system for detecting dangerous ice as small pieces ofglacial can often go undetected [5]. Here, we examine a method toaugment the radar system with image processing in the detection ofglacial ice.

spectral reflectance of glacialice, a classification system basedon k-means clustering wasimplemented. The figure on theright shows typical results with 7clusters with fresh snow anddirty glacial ice matching theexpected reflectance. Separatingthese further gave rise to resultsboth consistent with knownareas of glacial ice and the mostrobust results across all pictures.

NON-GLACIAL ICE

GLACIAL ICE

(left) The Arctic Cruise Ship  ‘Bremen’  (Source: Transport Canada)

LeWinter #82 – 80.3% repeatability with image #06