2021/05/11 by Özge Mercanoğlu Sincan, Sincan, Ozge Mercanoglu, Julio C. S. Jacques Junior +5 · 1 citation
Computer Science · Psychology · Engineering · #Hand Gesture Recognition Systems #Hearing Impairment and Communication #Gait Recognition and Analysis
paper · pdf · doi:10.48550/arxiv.2105.05066
The performances of Sign Language Recognition (SLR) systems have improved\nconsiderably in recent years. However, several open challenges still need to be\nsolved to allow SLR to be useful in practice. The research in the field is in\nits infancy in regards to the robustness of the models to a large diversity of\nsigns and signers, and to fairness of the models to performers from different\ndemographics. This work summarises the ChaLearn LAP Large Scale Signer\nIndependent Isolated SLR Challenge, organised at CVPR 2021 with the goal of\novercoming some of the aforementioned challenges. We analyse and discuss the\nchallenge design, top winning solutions and suggestions for future research.\nThe challenge attracted 132 participants in the RGB track and 59 in the\nRGB+Depth track, receiving more than 1.5K submissions in total. Participants\nwere evaluated using a new large-scale multi-modal Turkish Sign Language\n(AUTSL) dataset, consisting of 226 sign labels and 36,302 isolated sign video\nsamples performed by 43 different signers. Winning teams achieved more than 96%\nrecognition rate, and their approaches benefited from pose/hand/face\nestimation, transfer learning, external data, fusion/ensemble of modalities and\ndifferent strategies to model spatio-temporal information. However, methods\nstill fail to distinguish among very similar signs, in particular those sharing\nsimilar hand trajectories.\n