Equipment-free personnel action recognition and position estimation method based on multi-task learning

Abstract

The invention discloses an equipment-free person action recognition and position estimation method based on multi-task learning, and the method comprises an offline stage and a present stage, in the offline stage, an action label, an X-axis label and a Y-axis label are added to each CSI image to form a training set, then the training set is sent to a multi-task network for learning training, and a model is stored; in the online stage, the obtained CSI image is sent into the trained multi-task network model, and action recognition and position estimation are carried out. A multi-task deep neural network of a hard parameter sharing mechanism is utilized, only the correlativity of network parameters is researched in a backbone network, the correlativity and effect between tasks are emphasized, and the specificity of the tasks is ignored. And in the branch network, the structures are mutually independent, and the specificity of each task is reserved.

Description

technical field [0001] The invention relates to a multi-task learning-based non-equipment personnel action recognition and position estimation method, which can be used in the technical field of positioning and navigation. Background technique [0002] In terms of target state estimation, it is mainly to recognize the target's action and posture changes. Nowadays, there are many ways to detect and perceive these changes, such as wireless signals, wearable smart devices, image signals and video signals. Among them, most of the methods that require wearable devices use integrated devices with built-in wireless sensors. Although these devices are relatively cheap, they are inconvenient to wear on some occasions. However, the method of relying on images and videos has dead spots in shooting, and the shooting of images and videos is also easily troubled by light problems, which affects the accuracy of the final judgment. In addition, the methods of images and videos are also like...

AI Technical Summary

Problems solved by technology

And the position estimation problem is transformed into classification learning, which will lead to an increase in localization error when the target is located between the reference points

Images