PENILAIAN GERAKAN BARIS-BERBARIS BERBASIS AI DAN LSTM PADA SELEKSI PASKIBRAKA

Abstract

PASKIBRAKA member selection is conducted annually at three levels. PASKIBRAKA members must fulfil the criteria, one of which is the Marching Regulations (PBB). In the UN assessment process, sometimes the judges are still subjective. Therefore, it is necessary to have a system with artificial intelligence (AI) integration. The methods used in the system are Mediapipe and Long Short Term Memory (LSTM) algorithm. Mediapipe serves to display 33 keypoints and extract features from these keypoints which will later be processed by LSTM for motion detection. From the test results, the highest accuracy value of each movement is walking in place by 99%, striding by 95%, and regular steps by 97%. While the lowest accuracy value of each movement is walking in place by 12%, striding by 21%, and ordinary steps by 23%. The accuracy value is the result of the analysis of the LSTM model that has been made. The highest accuracy value in each movement comes from the correct movement, while the lowest accuracy value comes from the wrong movement.  The accuracy results in the recognition of stride and ordinary steps are influenced by video movements during the transition of the right hand to the left hand which are still detected incorrectly because almost all movements when both hands are in the lower position, causing low accuracy values.

Keywords: Keypoints, Long Short Term Memory, Marching, Mediapipe