Detect Specific Movement Patterns Based on Gyro and Accelerometer Data

Abstract: Wearable security cameras has been used by police officers and other agents working within the field of security for many years. Some of these cameras are equipped with an accelerometer and gyroscope sensor that can be used to detect if something is happening to the wearer of the camera. Ideally, the recording should start automatically if the camera is affected by sudden movements, for example if the wearer of the camera is being attacked. If this happens, then the camera should start recording immediately to capture the situation. This masters thesis project will be about trying to develop an efficient and accurate algorithm for fall detection with the help of the accelerometer and gyroscope sensors. Measurements and data collection was done by attaching a body worn camera to a user's chest with the help of a belt. The user performed different movements and the sensor output data was saved into separate files which was later made into a complete data set. Each file was then plotted for analysing, evaluation and trying to detect different patterns. An iterative strategy has been taken for developing this algorithm meaning that new functions and features were continuously added to the algorithm throughout this process in order to improve its performance. This Master's thesis project will focus on how to detect a fall while also trying to filter out movements and other actions that may look similar to a fall. The developed threshold-based algorithm was able to detect falls with a sensitivity of 100\% and a specificity around 90\%.