Bringing Computer Vision to Avalanche Detection

Last fall I had the chance to do an internship in the Alpine Remote Sensing group at SLF Davos, the snow and avalanche research institute. I had just completed two years of the Computer Science Bachelor and wanted to try something new. Throughout my internship, I worked on automatically detecting and classifying avalanches in RGB webcam images. This work aims at improving avalanche monitoring by leveraging computer vision techniques, making it easier to identify hazardous conditions in real time. (RGB cameras are designed specifically to capture visible light. They create images that replicate human vision, capturing light in red, green and blue wavelengths (RGB) for accurate color representation.)

Pia Ruttner-Jansen, who supervised me during my project at SLF, works on capturing fine-grained snow-depth distributions using automated measurement stations. This data is a valuable input for assessing avalanche risk. With higher precision, preventative measures such as road closures can be shorter, more targeted, and ultimately improve safety while reducing downtime. Each station is installed high in the mountains, directly facing a slope and equipped with LiDAR sensors. Although this type of sensor produces noisier data than commercial laser sensors, it is far more affordable, making it possible to deploy multiple stations at strategic locations.

Apart from a LiDAR sensor each measurement station also has an RGB camera that is set up to take a photo every hour. While not directly used to compute snow-depth distributions these images provide an additional, underutilised stream of data. The goal of my project was to use this data to automatically detect new avalanches and reduce the time spent manually clicking through images. In addition to learning about different types of avalanches and labeling images my work mainly involved experimenting with different computer vision approaches. Two major challenges were using a noisy dataset with only a small amount of unique avalanche events and choosing a suitable task that is useful for solving the problem while allowing for the use of already labeled data.

SLF has a truly unique working culture: when it first started snowing, the head of the institute sent out an email to celebrate this with free chocolate for everyone. Many PhD students have “Jochdienst” instead of teaching duty. In a weekly rotation, they take snow measurements up at Weissfluhjoch at 7:30 a.m. whatever the weather. Fascinatingly, it is the only place at this height on earth where there have been continuous daily measurements of fresh snow and snow depth for more than 80 years!

I found it very refreshing to experience a new environment. At ETH, I would usually mostly interact with other Computer Science students. Instead, the people I met at SLF had a background in Geomatics, Geology, Electrical or Mechanical Engineering, and worked on various problems such as mapping avalanches, studying avalanche dynamics or detecting cracks in borehole images. Everyone was super nice and helpful. The amazing cafeteria constituted a great way to chat with others and find out what they are working on.

ETH Studios

ETH Studios are global innovation outposts where ETH students tackle real-world challenges alongside companies, researchers, and startups. Whether in New York, Heilbronn, Davos, or at the experimental StudioX, students gain entrepreneurial and interdisciplinary experience while contributing to sustainable and tech-driven solutions.

Contact Francesca Caggiula at eth-studios@ethz.ch if you are interested in more information on and about the ETH Studios.