| The Internet of Things is a system that covers a large number of interconnected devices and sensors that can collect,exchange,and process massive amounts of data.As IoT technology continues to evolve and be put into practical use,user behavior analysis has become important in many fields,such as smart homes,smart transportation,smart health,and security monitoring.As the number of IoT devices increases and data processing capabilities improve,behavior analysis can provide users with more personalized and intelligent services.By deeply mining and analyzing user behavior data,the system can accurately understand user needs,optimize resource allocation,and improve service quality.In addition,user behavior analysis is also important in the security field,such as preventing and identifying security threats by analyzing abnormal behavior.However,existing behavior analysis systems still face many challenges,such as the difficulty of integrating and cleaning heterogeneous data,handling large amounts of data,and meeting real-time task requirements.To address these challenges,this study divides behavioral types in IoT environments into individual behavior and group mobility and proposes analysis methods for these two types of behavior using big data processing,machine learning,deep learning,and time-series analysis techniques.Specifically,this study proposes an identity authentication and home control method based on intelligent wearable devices,a real-time speaker recognition method based on edge-cloud collaboration,and a navigation data update method based on large-scale rider behavior.The main research content and contributions of this article can be summarized as follows:First,accurate and convenient user identification and individual activity understanding are the basis for analyzing user behavior.Therefore,this study proposes a system that relies only on small gestures to authenticate users and recognize activities and explores its application in the field of smart home control.Compared with traditional activity analysis systems,this system is based on commercial intelligent wearable devices and does not require additional hardware,making it more user-friendly.The system design includes an inertial sensor data preprocessing method and a deep temporal neural network for accurate user and gesture recognition.Experimental results show that the system has high accuracy and reliability and brings new gesture recognition methods to the field of smart home control,which is expected to promote the development of related technologies.Secondly,in practical applications,the real-time performance of behavior recognition tasks is difficult to meet the requirements due to factors such as node resource limitations,network transmission pressure,and high costs.Therefore,this paper proposes a framework for collaborative edge and cloud computing to detect active speakers in real-time under the camera.The framework uses the latest developed ASD model based on facial movements and a voiceprint generator to accurately detect active speakers in noisy and crowded environments.At the same time,this paper also studies the effectiveness of a series of audio and visual features as filtering clues under the lens and provides two real-time collaboration frameworks for different edge resource limitations.Through a large number of experiments on real datasets,the effectiveness of the proposed method in identification accuracy and filtering efficiency has been verified.Finally,a deep understanding of group mobility is a key link in analyzing user behavior in IoT systems.Therefore,this paper proposes a method based on abstract graph analysis of group mobility data for analyzing user behavior in IoT systems.The method uses dynamic abstract graphs to maintain navigation data and queries external map servers to ensure the accuracy of cached paths.In addition,in the on-demand delivery scenario,the method uses courier data for behavior analysis and adopts a query algorithm based on UCB to update database storage for real-time access to traffic information.The proposed method was deployed in the database of one of the largest on-demand food delivery platforms in China and its performance was evaluated against the state-of-the-art methods.The experimental results show that the proposed method achieves significant savings in the required number of queries and improves the accuracy of cached paths. |
