Real-time, fine-resolution human intra-gait pattern recognition based on deep learning models

A method of classifying gait patterns from images, comprising:

• capturing, by an image capture mechanism, one or more images of a gait-cycle of a person;
• and
• receiving, by a computing device, the captured one or more images, said computing device comprising at least a processor in communication with a memory, wherein computer-executable instructions are stored on the memory that cause the processor to:
• classify one or more gait patterns using deep learning on the captured one or more images in real time using skeleton-based gait recognition,
• wherein the skeleton-based gait recognition comprises estimating joint locations and limbs of a human body from the captured one or more images of the gait-cycle of the person using a pre-trained pose estimation framework, said pre-trained pose estimation framework comprising:
• feeding the captured one or more images into a trained convolutional neural network (CNN) to perform predictions of the joint locations;
• using confidence vectors to identify the joint locations;
• associating identified joint locations according to a confidence map;
• and
• bipartite matching to connect the identified joint locations;
• wherein multiple consecutive images are used to construct a discriminative feature for classification of the one or more gait patterns;
• and
• wherein the multiple consecutive images comprise a first image and a second image, each of the first image and the second image comprising of six joints of a lower body of the person are collected in each of the two consecutive images and a previous frame is subtracted from a latter frame to construct a temporal feature, wherein the temporal feature provides the discriminative feature for classification of the one or more gait patterns.