Recognition of Similar Gait Pattern Using Transfer Learning DarkNet

Abstract

Nowadays, convolutional neural network as a recognition model has improved human identification by gait patterns. The model recognizes personal steps of walking and its performance is up to a training dataset. After a pre-processing for human segmentation, it shows that the gaits for each person are quite similar with no significant difference. This causes a low accuracy in the recognition. To solve this issue, this paper proposes DarkNet-19 model to improve human detection and is adapted to classify gait pattern by comparing in pixel levels. A new model is developed based on CASIA-B dataset. In the experimental results, a recognition rate and identification accuracy are increased. The proposed method enhances an ability to identify person, who is walking, carrying a backpack, and wearing a coat, of 99.83%, 98.10% and 99.85% respectively.