Identification of personnel identity in unmanned aerial vehicle video gait recognition based on 3Dresidual network

An improved adaptive 3D ResNet-18 residual network gait recognition method is proposed for the problem of recognizing people's identity in videos captured by UAVs. Using a combination of gait feature extraction and deep learning models, human silhouettes are separated from complex backgrounds for modeling, and their gait features are extracted for identification.3D ResNet-18 residual network learns spatio-temporal features to extract temporal and spatial features in the sequence of video frames and introduces a cross-entropy loss function to optimize the model and improve the prediction accuracy. When dealing with the challenges of motion blurring in the video captured by drones and shadows and noise interference in background modeling, the improved MOG2 background separation algorithm detects pedestrian gait and models it, and adaptive variance thresholding improves the background detection capability and improves the spatio-temporal feature extraction and optimization, which significantly improves the robustness of the model. The experiments use the Walking category in the KTH public dataset, and the results show that the model exhibits good recognition results in different backgrounds. This approach effectively improves the performance of UAVs in applications such as security surveillance.