Human Recognition Through Gait Audio

Human recognition through gait has always been a challenging field and to recognize a person through gait audio samples is even more challenging and exciting. In this present research, gait audio samples are utilized for person recognition in three different scenarios-normal, with a bag pack, and with shoes on. Gender estimation is also attempted and the results are quite satisfactory in comparison to existing techniques. In this work, gait audio samples are converted to image data specifically spectral representation, and fed to a convolution neural network for further processing. RESNET34 model is particularly utilized as a deep network. Gait Audio Dataset from TUM repository is used for this research. Recognition results reveal that the proposed model is quite better in terms of accuracy and provides motivation to explore CNN models for audio data.