Human Behavior Recognition For Home Service Robot

In the area of research on home service robots,how to make robots accurately and quickly recognize human behaviors has become one of the important research topics in this field.Service robots can provide better services to humans if it has the ability to identify the service object.This dissertation comprehensively analyzes the relatively mature methods of human behavior recognition in recent years and comprehensively considers the particularity of human behavior in home service robot platform and home environment.Combining with the depth image processing method,this dissertation proposes a new human behavior recognition for home service robot,and focus on analysis and research from the following two aspects.Then,the point cloud information of the fusion depth image is used as the fourth feature,and these feature values are taken as the feature vectors of the dynamic behavior.First of all,determine the behavior of joints to complete the extraction of human behaviors based on the characteristics of human behaviors.The Kinect sensor is used to acquire the skeletal model and depth image.We can get the orientation matrix and the three-dimensional coordinates of bone,which extracted from Kinect sensor.Select the key bone joint point about dynamic behaviors,and the body posture features,hand position characteristics,and motion information are obtained through the processing of the data information which are provided by the skeletal model.Finally,the template matching behavior identification algorithm is used to complete the identification.According to the similarity between the key action sequence and the key action sequence in the template,the behavior category is determined,and the corresponding behavior category is the final result of the recognized human behavior.Secondly,the human behavior recognition method is studied for the extracted features.Based on the idea that a behavior is a key action sequence,and then the human behavior recognition model is constructed.Using the Gaussian mixture model to determine the key action candidate set,and through the entropy model to select the key actions that constitute the human behaviors.Next,using the improved genetic algorithm,whose fitness function is the minimum edit distance,to complete the operation and optimize the key action sequence template.Finally,the 12 human behaviors that often appear in the home environment are selected to identify the human behavior.The recognition experiment is performed from the mirror action and the light intensity,and then the analysis results are summarized.It proves the feasibility of the method proposed in this paper in real life.