Person Re-ID is defined as a matching problem of multiple instances of a person captured by different cameras (i.e. different views). Depending on the number of images used for person representation, person ReID methods can be categorized into two approaches: single-shot (in which one sole image is used) and multi-shot ones (in which a sequence of images is used for each person). Person Re-ID is non-trivial problem due to different challenges such as strong variations in illumi- nations, poses, view-points and visual appearance; occlusion happened in crowded scenes. In order to overcome these challenges, many great efforts have been made to build robust descriptors for person representation [3], to propose effective distance metrics for person matching or to combine different features through fusion technique
Exploiting matching local information for person re-identification (2022)
The use of hand gestures provides an attractive alternative to cumbersome interface devices in human-computer interaction (HCI). In such applications, gestures are acquired by sensors then automatically recognized and matched to several commands to control the machines. In the literature, a number of studies on this topic have significantly increased in recent years due to their wide applications in virtual reality, game, and healthcare or HCI. To be deployed in practice, hand gestures need to be intuitively designed while still being easy to memorize for users, easy to implement, and simple to deploy. Besides, the algorithms for hand gesture recognition must be highly accurate and lightweight to run on low-resource computers or edge devices
Hand Gesture Recognition From Wrist-Worn Camera for Human–Machine Interaction(2023)
Dynamic Hand Gesture Recognition from Egocentric Videos based on SlowFast Architecture(2022)