To search, Click below search items.


All Published Papers Search Service


Reliability of Low Framerate in Deep Learning Based Visual Human Gait Analysis for Identification


Salisu Ibrahim Yusuf, Steve Adeshina, Moussa Mahamat Boukar


Vol. 22  No. 1  pp. 411-418


Extracting images from video dataset determines input and computation demand in training a machine learning model, using less framerate cut down on cost, however, higher framerate leads a better video quality hence better result is expected. In this paper, we investigated the influence of framerate on accuracy in identification of human through gait using the convolutional recurrent neural network model. Primary dataset was collected from at 60fps for 26 persons each having 12 to 16 instances. Frames are extracted at varying rates 60, 30, and 15 fps. We measured and comparatively analyze the accuracy and loss in train and validation set using each instance of datasets on the same model. The outcome shows similar results for each framerate and slightly better validation accuracy was recorded for 15fps


computer vision, human identification, framerate, deep learning, video dataset preprocessing, CNN-LSTM, gait analysis.