Intelligent Systems Lab Project: Vital Status Monitoring of Elderly People in an Intelligent Room
Participants
- Niels Dehio
- Denis John
- Christian Menßen
Project Supervisors
- Peter Christ
Motivation
- Elderly people often get insufficient exercise in their daily life
- In normal room environments, the amount of sportive activities cannot be determined
- An intelligent room could help to solve this problem by monitoring the user, equipped with vital sensors, to determine the actual activity
- Users (or their doctors) could particularly profit from daily reports of activities as they can help to detect changes early. Furthermore, the training progress could be monitored with long term statistics.
Application Szenario
In a typical situation in our intelligent room, the elderly will be equipped with two body sensors, which are attached around the chest and the wrist with a strap. The body sensor integrates a three dimensional acceleration sensor and a heart rate detector, developed by Cognitronics and Sensor Systems Group, Bielefeld University. These two body sensors are used to record the activities of a person throughout the day.We developed a system with the help of these sensors which allows therapists to regularly analyze the effort of their patients and motivate them to do the required amount of sportive activities. Another feature of our system is the evaluation of the current activity. A website displays the current activity live.
Objectives
The project goals are- to capture vital parameters with two vital sensors
- classification of the actual activity with machine learning algorithms
- cumulation of the recorded activities to daily statistics and publish them on a website
- to display the current activity on the website
Description
Requirements
- Two vital sensors with corresponding USB antenna
- Installed RSB / RST 0.7
- Apache Ant
- Unix operating system (Ubuntu 12.04 for example)
- Java 6 and Weka 3.6.9
- MySQL Server with imported database schema
- Apache webserver
Prepare for usage
Place one sensor in the bracelet and put it on the right arm as shown in the picture:
The second sensor should be placed at heart height with the provided chest strap.
System architecture
VitalClassifier: Written in Java, listens on the RSB scope /vital. Classifies the incoming data with a J48 decision tree and stores the classification result in a MySQL database.
Website: Written in PHP, displays the current activity and generates cumulated daily and monthly statistics of the executed activities.
Results
- Our results are based on 20 minutes of labeled training data per person.
- After filtering the training data, our classification algorithm is able to distinguish 10 different activity classes
- The features considered in this work are based on the publication: Activity Recognition using Cell Phone Accelerometers
- Within a 10-fold cross-validation, an accuracy of 78% was achieved using a decision tree (J48)
Discussion and Conclusion
- The planned results have been achieved.
- Because of RSB, it is easily possible to share the sensor values with other applications in the intelligent room.
- With respect to our testing process, we proclaim a high generalisation among different users.
Outlook
There are many possible new features to be implemented:- More activities should be detectable
- For a longer battery life of the vital sensors, the feature calculation can be implemented on the micro controller.
- To simplify the process of training, a guided training application should be available.
- In case of emergency, the system should able to automatically alarm someone. For example, these cases could be a fall or a heart attack.
- In general, it is possible to replace the vital sensors with an Android smartphone. The quality of these results has to be analyzed.