26th IEEE Signal Processing and Communications Applications Conference (SIU), İzmir, Türkiye, 2 - 05 Mayıs 2018
Smart phone use in indoor positioning systems are popular. In the case of commonly used received signal strength (RSS) based positioning methods, the problem is that RSS values vary with time due to radio frequency signal characteristics. Human presence, including user/device orientation, decrease positioning accuracy, with significant influence on signal characteristics. Updating the RSS values used in positioning algorithms based on fingerprint or geometric methods is a good solution to increase accuracy. In this study, an intermediate solution is proposed that recognizes the location of the smartphone on user. Log-Sigmoid function and Softmax regression are selected as activation functions with cross entropy measure and artificial neural networks are used for classification purpose. Using inertial sensor data of the smartphone, 9 target classes are successfully classified. Over 95% success rate is achieved for 50 runs within generated dataset.