Activity Recognition using Context Awareness

Our prime customer was looking for Context-as-a-Service, where context can be sensed, thoroughly understood and based on that further actions can be performed. Context involves – Identity Context (Who am I?), Activity Context (What I am doing?), Time Context (When I am doing?), Location Context (Where I am?). Context-Aware systems can sense their physical environment & adapt their behavior accordingly. A requirement was to develop a library that can sense human activity (standing, walking, sitting, running) and collect data for the user’s current activity and based on the collected data provide the recommendations.

  • Design and development of Context-Aware Library Framework which is modular, extensible to include various context-aware algorithms and portable on any platform. These algorithms should be accessible to application layer through easy to use APIs
  • Design and development of Activity Recognition Algorithm based on accelerometer sensor input and runs as part of context-aware library

There were few of the following challenges that we came across developing this system more accurate:

  • Collecting data for user activity as each individual has different pattern of doing any activity
  • As working environment can impact the user pattern of doing an activity, it was a challenge to write an algorithm for collecting and analyzing data for similar kind of activity in different environments
  • As each user has different speed of performing an activity, it was also a challenge to decide how frequent a system should collect the data
  • Highly Modular & Extensible Context-Aware Library written in C
    Capable of integrating multiple context-aware algorithms and sensors easily
  • Set of intuitive APIs available which allows easy integration
  • Low memory footprint
  • Portable on any platform easily
  • Supported Sensors: Accelerometer, Gyroscope, Magnetometer, Microphone, Humidity, Pressure, IR temperature sensors
  • Supported Algorithms: Activity recognition, Posture detection, Sleep detection, Fall & Step detection
  • Support for basic activity classification like Resting, Walking, Running, Stair Climbing (up or down), Standing, Fall Detection
  • Library sends the classified output on wearable device display
  • Accuracy greater than 85% achieved
  • Takes accelerometer data as input
  • Algorithm can be optimized for any user’s position/orientations
  • Designed & Developed Context-Aware Library Framework
  • Designed & Developed Algorithms for Activity Classification, Posture detection, Sleep detection, Fall & Step detection
  • Porting the Library and Algorithm on microcontroller based wearable platform from TI