Annotation tool designed for hazardous user behavior in guided mountain transport

This paper proposes a semi-automatic ground truth annotation software designed for the specific needs of the EVEREST project. The purpose of this project is to build an annotated and anonymized video database, and use it to evaluate algorithms in the task of detecting hazardous behavior in guided mountain transport. To do so, a ground truth annotation tool that disposes designed specifically for the EVEREST project was needed. Ski lifts safety based on intelligent video systems is a niche domain which has not yet been explored in depth, which means no annotation tool suited for this task was available. That is why, we decided to develop a user-friendly and flexible tool to allows the semi-automatic annotation of events and faces (for privacy purposes). We looked at existing tracking algorithms, chose an implementation of TLD, and designed a new tracking algorithm that could be used when TLD isn't effective. This led to a simple, lightweight tracking algorithm that is more practical to use than the original CAMshift algorithm, and a user-friendly and flexible annotation tool that is well adapted to the specific task of annotating hazardous behavior in guided mountain transport.