Research On Human Motion Intention Recognition Method Based On Police Patrol Robot

As a mobile robot,the police patrol robot was designed to partially assist or completely replace police officers to carry on daily security work,and its intelligent monitoring and proximity detection capabilities can significantly improve security quality.Using surveillance video to realize human behavior recognition was an important task that police patrol robots need to complete in the process of security patrol This paper proposed new human behavior recognition methods from the perspective of RGB video and skeleton sequence data,and designed and completed a human behavior recognition system based on the police patrol robot experimental platform.The specific work was summarized as follows:A human behavior recognition algorithm based on dense three-dimensional residual network was proposed.In order to solve the problem that the existing algorithms could not make full use of the multi-level spatio-temporal information of the network,a new human behavior recognition method based on RGB video was studied.First,the three-dimensional residual dense block was used as the basic building block of the network,and the densely connected convolutional layer was proposed to extract the hierarchical features of human behavior.Second,the local dense features of human behavior were adaptively learned through local feature aggregation.Thirdly,the residual connection module was proposed to promote the flow of feature information and reduce the difficulty of training.Finally,the multi-layer local feature extraction of the network was realized by cascading multiple three-dimensional residual dense blocks,and the global feature aggregation adaptive method was adopted to learn the features of all network layers to realize human behavior recognition.Experimental results on various data sets show that the proposed algorithm architecture had good robustness and transfer learning ability,and could effectively handle a variety of human behavior recognition tasks.A human behavior recognition algorithm based on local feature fusion was proposed.Considering the robustness of the human behavior recognition algorithm in complex scenes,this paper proposed a human behavior recognition method based on skeleton data.The local features of skeleton relative position,joint angle,joint point speed and body speed were extracted from the skeleton sequence through improved local feature extraction,and were fused to describe different types of actions.The designed feature fusion method well described the spatial static information and temporal dynamic information of the action,and also used Principal Component Analysis to reduce the dimension and noise,and saved the feature information with strong recognition ability.Finally,the paper verified the effectiveness of the proposed algorithm on the data set built by ourselves.Based on the above algorithm,based on the police patrol robot platform,the paper designed a human motion behavior intention recognition system,which could complete the patrol task in a fixed area,and realize real-time video monitoring,human behavior tracking and recognition during the patrol process.After testing in actual scenarios,the effectiveness of the system was proved.