Research On Police Situation Recognition And Police Situation Classification System Based On Police Patrol Robot

Police patrol robots are mobile robots that integrate autonomous navigation functions and typical police situation recognition functions.They play patrol and security roles in some public places and major competition areas.From the perspectives of spatiotemporal graph convolutional neural network and spatial graph convolution theory,this paper proposes STGCN(Spatial Temporal Graph Convolutional Networks)human action recognition algorithm based on new spatial partition and ST-GCN based on symmetric normalization Human Behavior Recognition Algorithms.Based on the police patrol robot platform,the police situation recognition system and the police situation classification system are designed.The specific work is as follows:1)A new space partition-based ST-GCN human action recognition algorithm is proposed.This algorithm is aimed at the problem that the root node and the farther nodes cannot establish a connection relationship in the original space partition of the ST-GCN model,which leads to the problem that the convolution operator cannot generate a higher-level feature map for the local features of the human body.The new spatial partitioning strategy improves on the original spatial partitioning: further expands the range of the neighborhood set and calculates the distance between each node and the root node to cluster each node into 4 subsets.In addition,at each layer of the ST-GCN model Add a learnable parameter mask M and residual module,and finally verify the algorithm through public datasets.2)A ST-GCN human action recognition algorithm based on symmetric normalization is proposed.For the original ST-GCN model,the normalization method aggregates node features,which has the problem of incomplete scaling of highly neighbor nodes.The symmetric normalization strategy uses two degree matrices to normalize the rows and columns of the adjacency matrix at the same time.The purpose is to attach greater weights to low-degree nodes to reduce the influence of high-degree nodes on the whole.Finally,the symmetric normalization is verified by public datasets.Unified aggregation algorithm.3)Based on the police patrol robot platform and the previous improved algorithm work,the police situation recognition system and the police situation classification system are designed.During the patrol of the scientific research building,the police situation recognition system was tested outdoors and the police situation classification system was tested indoors,which proved the effectiveness of the system.Finally,the thesis shows the visual interface of police situation recognition and police situation classification results.