GuoQing Yin

Daily activity learning from motion detector data for Ambient Assisted Living

In an intelligent environment one important task is to observe and analyze persons daily activities. Through analyzing the corresponding time series sensor data the persons daily activity model should be build. To build such a model some problems have to be overcome: the sensor data count increase sharp with time and the distribution of the data is dynamically according the persons daily activities. In an Ambient Assisted Living (AAL) project we handle this kind of time series sensor data from a motion detector. At first we reduce the data count through a predefined threshold value and build data “states” in time interval. Secondly, we analyze the states using a hidden Markov model, the forward algorithm, and the Viterbi Algorithm to build the persons daily activity model. To test the correctness of the model some special and random days activities routine will be given.

Build user daily activity model and model structure changing

Because of the activity dynamic it is a challenged work to build daily activity model of user. The goal of the paper is to build daily activity model of the user with high accuracy. At first the raw data derived from motion detector will be “translated” to state data, then use state split and merge to build the basic model. Thirdly in order to increase accuracy of the model the count of the state data increased by changing the parameter of the translator. Here we get another two activity models with higher accuracy. The changing of the count of the state data caused to the structure changing of the build activity model. We merge the states from same route of activity models, compare different models and find out the same activity trends of the user. On the other hand from the models with higher accuracy the hidden activity will be detected.

Data analyzing and daily activity learning with hidden Markov model

To observe and analyze persons daily activities, and build the activities model is an important task in an intelligent environment.

Gaussian Mixture Models and Split-Merge Algorithm for parameter analysis of tracked video objects

Parameters of tracked video objects (for example: the angles of moving objects) are discrete random variables and the amount of data increases over time. In this paper we use a new method to analyze the parameter angle: the video frame is segmented into small sections and in each section the angle values during some time period are gathered. Through analysis the angle data in each section these angles can be modeled, therefore also in whole frame. The build model will be used to find abnormal behavior of moving objects. To build a statistical model of the angle of moving objects from the video data is a question of cluster analysis in real time. For this application, Gaussian Mixture Models and Split-Merge Algorithm provide a powerful solution.

Daily Activity Model for Ambient Assisted Living

We propose a novel way for ambient assisted living: a system that with motion detector to observe the daily activities of the elderly, build the daily activity model of the user. In case of unusual activities the system send alarm signal to caregiver. The problems with this approach to build such a model: firstly, the activities of the user are random and dynamic distributed, that means the related data is dynamically and with huge count. Secondly, the difficulty and computational burden to get character parameters of hidden Markov model with many “states”. To deal with the first problem we take advantage of an easy filter algorithm and translate the huge dynamical data to state” data. Secondly according the limited output of distinct observation symbols per state, we reduced the work to research the observation symbol probability distribution. Furthermore the forward algorithm used to calculate the probability of observed sequence according the build model.

Split-merge algorithm and Gaussian mixture models for AAL

Analyzing time series sensor data and build statistical model in real time has to overcome two problems at least: the data count increase with time and the distribution of the data is dynamically. To deal with this kind of problems Gaussian mixture model and split-merge algorithm provide useful way. In an AAL project we handle the time series sensor data from a medical box contactor and a meal entrance contactor. Using Gaussian mixture model and split-merge algorithm to analyze the sensor data gathered for about one and a half months and built the statistical model.

Activity Analyzing with Multisensor Data Correlation

In an ambient assisted living project, a novel way to be proposed in order to protect privacy, increase comfort and safety: a system that with different kinds of sensors installed in the living environment and observe the daily activities of the elderly. Based on the daily activities of the user an activity model will be build. In case of unusual activities the system will send alarm signal to caregiver according the build activity model.

Fast learning algorithm for Gaussian models to analyze video objects with parameter size

Size of objects in scenes is an important parameter of video surveillance systems. From the analysis of objects size we can build an objects size model in scenes. The basic idea derives from automatic calibration to different perspectives. To build such a model of objects size from the real time video data we utilize Gaussians and real-time fast learning algorithm from literature. The built model is used for real-time surveillance systems.