A new method for activity recognition using smartphones is proposed. Using three-axes accelerometer and gyroscope signals, the proposed system is able to identify low level activities with a high level of accuracy. The method works regardless of orientation of the device with respect to the body part to which it is attached. The algorithm achieves a high level of accuracy when trained on a small set of users and tested on an unknown user.
Recently indoor positioning methods based on WLAN signal measurements gained popularity because of high localization accuracy. These methods exploit radio maps obtained from wireless signal measurement surveys on location grids. Measurement sets from various WLAN access points are called fingerprints and characterize locations where the measurements are collected. As WLAN environments do not ensure continuous measurements availability, and faulty or rogue access points may unexpectedly change surveyed signal patterns, resiliency becomes an important issue to address using algorithmic methods. This paper first proposes a general fault model that integrates several reported models. Then performance degradations due to faults are studied for conventional fingerprinting methods. Two improvements to positioning systems are proposed for mitigating the impact of faulty measurements. The first improvement takes into account the intermittent unavailability of AP samples when calculating kNN. The second improvement allows the system to switch from a high accuracy method that works only under normal conditions, to a more resilient method whenever a high number of faults are suspected. Performance figures are provided for positioning with data surveyed from a real environment, to which varying amounts of faults have been introduced artificially.