Human Behavior Understanding Based On Semantic Analysis In Home Intelligent Space

To realize human-computer interaction and independent service of service robots,it is an important basis to understand human behavior in a complex family environment.For research of human behaviour understanding in the family environment,most of them are only limited in the detection of sudden abnormal behavior and abnormal habits.Methods of common human behavior recognition are more suitable for general structured environment.When these methods are directly to the unstructured family environment,it will cause the problems of complex system modeling,inconvenient information collection,unitary behavior recognition.Considering to taking full use of human interaction,this paper combines object recognition and action recognition and use ontology as a vehicle.For human behavior understanding at home,it presents a method of human behavior understanding based on semantic analysis in home intelligent space to solve these problems.Aiming at realizing the real-time information acquisition and full use of service object to service robot,we construct the human ontology library in home intelligent space.The relevant information of service objects is represented in user field and environment field,we choose to use relational database persistence as ontology storage methods.It's not only to realize the real-time information acquisition and full use of service object but also to query in the knowledge of ontology efficiently.It lays the foundation for the realization of human behavior understanding based on semantic analysis in home intelligent space.In order to recognize human motion more accurately in complex family environment,a method of human action recognition based on residual neural network at home is proposed.Its innovation lies in dataset's pre-processing and residual neural network's application.This approach extracts foreground by background subtraction algorithm based on mixture Gaussian.Then it trains human action recognition model according to learning action features of datasets automatically by residual neural network.Because of the pretreatment for dataset and the use of better optimized residual neural network structure for the deeper network,this method can achieve better recognition results.When the current methods are carried out for better performance of depth model in image classification,it performs the problem of lack of hardware performance,difficulty in structural innovation and limited training samples.Considering to information fusion from different learning network models,an object fusion recognition algorithm based on DSmT(Desert-Smarandache theory)is proposed.Its innovation lies in combining the classic deep learning model and DSmT theory.The procedure of object recognition could be completed by the DSmT combination theory in the decision-layer fusion,the method of threshold contrast is used to judge the result of fusion.Conclusions of the work prove that this algorithm could improve correct recognition ratio effectively under the same conditions.A method of understanding of human behavior based on semantic reasoning in home intelligent space is proposed.Its innovation is to combine action recognition and object recognition and use ontology as a vehicle to achieve human behavior understanding on semantic level.JESS is used to match human behavior ontology library with human behaviour rule library,it generates multiple phrases.We achieve complex human behavior understanding on semantic level.Compared with other methods,the method based on ontology modeling is relatively simple,it's also convenient for information acquisition because of relying on intelligent space,and it focuses more on human-computer interaction with a certain ability of scene recognition,it is suitable for understanding of single person behavior in complex family environment.All in all,it can recognize and understand multifarious human behaviour in key frames accurately.