Gait Identification using Deep Convolutional Network and Attention Technique

Gait recognition has several merits, including the fact that it is non-invasive and simple to identify from a considerable distance. Gait patterns are commonly utilized to superimpose and synthesize pictures and generate energy-like templates in several commonly used picture or video-based gait detection systems for multi-view scenarios. However, because gait detection is sensitive to sample collection, information could be lost throughout the images compositing process. This paper presents a multiple view gait detection method using deep convolutional neural networks (CNN) and a channel-based attention technique combined to more successfully handle the aforementioned problems. To begin, the pre-processing method generates gait energy pictures as input to the neural network. Then training of each convolving layer using the back-propagation learning method, which increases the CNN’s learning capabilities. Finally, the neural network’s channel attention method is incorporated, which improves the capacity to convey gait characteristics. Softmax classifier function is used to categorise gaits. According to experimental results, the suggested technique outperforms many other contemporary CNN methods in terms of recognition effectiveness, by using the open-access gait data sets i.e. CASIA-B. Therefore, the convolutional neural networks and channel attention mechanism combination is quite useful for recognizing gaits.