DeepEvent is an application based on recurrent neural networks, which has been developed for the automatic detection of walking events. The network uses the 3D position and velocity of markers placed on the lateral malleolus, calcaneus, and second metatarsal to estimate the pose (PP) and detachment (DP) of the foot. The method was developed using 10526 PP and 9375 DP from 226 children with motor disabilities. The prediction error is 5.5ms for PP and 10.7ms for DP. DeepEvent is more accurate than trajectory-based algorithms usually used in motion analysis.
Télécharger le code en open source
En savoir plus sur la méthode de DeepEvent