Validade, confiabilidade e acurácia do Microsoft Kinect para registro de sessão de treinamento de taekwondo

• Ano de defesa: 2015
• Autor(a) principal: Cristiano Arruda Gomes Flor
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Kinect; Acurácia; Validade; Carga de treinamento; Reconhecimento de movimentos; Taekwondo; Confiabilidade; Artes marciais; Biomecânica; Tae-Ken-do
• Link de acesso: http://hdl.handle.net/1843/BUOS-AQWHEL

This study aims to determine the validity and reliability in getting the run time of a Taekwondo kick, and to verify the accuracy of automatically recognize movements inherent in a Taekwondo training session of a system comprises for the Microsoft Kinect sensor and algorithms developed specifically for these tasks. The study consisted of two stages. In the first step the volunteers performed 15 repetitions of Taekwondo front kick (ap chagui) in two different running speeds. To verify the concurrent validity, it calculated the Pearson correlation coefficient, the equation of linear regression and standard error of the estimate (SEE), checking the validity between the Microsoft Kinect and Optitrack (gold standard). To check the reliability were calculated the intraclass correlation coefficient (ICC2,1) and the Standard Error of Measurement (SEM). In the second phase, the volunteers were four different movements of a Taekwondo training session for 20 seconds each and was checked on the system's ability to recognize and automatically classifying movements which were conducted with high accuracy. The kick runtime system showed high reliability (ICC2,1=0.971 for slow kicks and 0.912 for fast kicks kicks). It also presented a high validity for kick speed 3 ms-1, with r=0.982 (p=0.001), a percentage of 96.5% of the model explained, linear coefficient 'a'=0.0174, angular coefficient 'b'=0.984 and SEE=0.064 seconds; and also for kick speed 8 ms-1, with r=0.920 (p=0.001), a percentage of 84.7% of the model explained, linear coefficient 'a'=0.336, angular coefficient 'b'=0.866 and SEE=0.011 seconds. In recognition of the movement accuracy mean was 89,91%, the right punch movement with more accurate recognition (93.81%) and the left punch with the lowest recognition accuracy (85.23%). The evaluated system is presented as a viable way to develop a Taekwondo training sessions record tool.