Study of the combination of modalities and coherence for the ReID of participants in Ultra-Running Scenario.
In the context of ultra-running, verification of participant's identities is necessary in order to avoid cheating and irregularities, and for safety and regulatory reasons the position of the runners must be known. In this context, state-of-the-art whole-body based re-identification approaches have reported moderate success due to challenging scenario characteristics such as: lighting or clothing changes. In this project, we explored different deep learning and machine learning techniques applied to ReID using as features: the body, the face and a combination of both. In addition, we investigate the use of contextual information to reduce the gallery size, and therefore improve the overall re-identification performance.