Probabilistic Soft Logic Based Human Activity Recognition Method In Smart Home

Human activity recognition technology refers to the process of inferring the user's intention automatically by monitoring a user's action and its located environment,and is widely used in many fields such as public safety and smart medical treatment.In S-mart Home environment,ADL are often characterized by complexity and uncertainty.Markov Logic Networks incorporates probability graph model and first-order logic,which is the perfect strategy to solve the complexity and uncertainty problems.However,in MLN-based activity recognition method,any atomic event adopts a hard constraint with a Boolean value,so that its expression ability for uncertain events is limited.In addition,the problem of Maximum posteriori probability inference in MLN is the one of integer Linear programming,which is hard to converge to the optimal solution,so both the recog-nition accuracy and the inference efficiency of MLN cannot service the request of human activity recognition technology.Aiming at above-mentioned deficiency of MLN method,in this paper,based on Prob-abilistic Soft Logic,an innovative method of human activity recognition is proposed.In contrast to MLN,which can only represent discrete variables,PSL adopts soft constraints on atomic events to enhance the model's expression ability for uncertain events.More-over,PSL uses Lukasiewicz logic instead of Boolean logic to relax the range of feature functions to interval[0,1],such that the integer linear programming problem is trans-formed into convex optimization problem,so as to improve inference accuracy while en-hancing inference efficiency.The contributions of this paper are as follows:According to the uncertain characteristics of ADL,a PSL-based human activity recognition framework is proposed.It not only can define uncertain logical relations,but also can provide an expression mechanism for uncertain sensor events,so that a novel uncertainty calculating method for sensor events is proposed based on DS evidence theory.In view of the temporal complexity of ADL,an activity modeling method named as PSL-EC combined with Event Calculus and PSL is proposed.On the basis of PSL,this method can describe the persistence of activities using calculus axioms,such that the PSL's ability for expressing complex temporal relationships can be expanded.