End-to-end human action recognition method, equipment and medium from the perspective of UAV

Abstract

This application discloses an end-to-end personnel action recognition method, device and medium from the perspective of an unmanned aerial vehicle, including: constructing and training a personnel action recognition network model; the model includes a feature extraction network, a personnel target detection sub-network, and a multi-target tracking sub-network Network and human action recognition sub-network; input the image to be tested into the model, use the feature extraction network to perform feature extraction, and the extracted feature map is shared by the three sub-networks; use the human target detection sub-network to detect the current frame with a bounding box The target in the target; use the multi-target tracking sub-network to perform inter-frame multi-target tracking according to the target's appearance feature vector and bounding box; use the human action recognition sub-network to integrate the motion information of the same target between different frames, and identify the target in the current frame type of action. In this way, the three sub-tasks of detection, tracking, and recognition are integrated into the same neural network, avoiding a large number of repetitive feature extraction and calculation redundancy, and achieving the effect of real-time behavior recognition.

Description

technical field [0001] The present invention relates to the technical field of action recognition, in particular to an end-to-end human action recognition method, device and medium from the perspective of a drone. Background technique [0002] Personnel action recognition technology is the key technology of the intelligent monitoring and analysis system. Combined with advanced drones and high-definition cameras, it can form a long-distance cruise warning system and enhance reconnaissance and counterattack capabilities. This is a space-time sequence action positioning task. It is necessary to locate the location where the person's action occurs in each frame of video, and at the same time determine the start and end time of the action. [0003] With the improvement of hardware performance and the application of GPU accelerated computing, methods based on deep learning have achieved great success in the field of computer vision. At present, in order to ensure the accuracy of ...

AI Technical Summary

Problems solved by technology

At the same time, the change of the shooting angle of the drone will bring about the relative change of the appearance of the personnel in the picture, which also affects the effect of personnel motion recognition.

Images