Research And Application Of User Behavior Based On Smart Phone Built-in Sensors

In recent years,smart phones can access to user's behavior habits due to smart phones with a variety of sensors. With the rapid development of user behavior research in the field of human-computer interaction, it has become a new trend to study user behavior with smart phone built-in sensors. Therefore, we use android phone to study user behavior which contains fall behavior detection and signature behavior verification in this work. The specific research works are summarized as follows:1) Research of fall behavior detection algorithmFall is the leading cause of accidental injury in the elderly. This is the reason,fast and early detection of the fall is very important to save and rescue the people and avoid the badly prognosis. In order to detect fall behavior, we are presenting athreshold-based fall detection algorithm in this article. The algorithm uses Signal Vector Magnitude (SVM) peak value, base length, post-impact velocity, residual movement and vertical acceleration to distinguish falls from most of daily activities.In the experiment, Activities of Daily Life (ADL) and fall data are captured using triaxial accelerometer and magnetometer of Android phone. The final experiment includes data from 120 simulated falls and 150 daily activities. Compared with previous methods, the proposed method achieves higher sensitivity and specificity.2) Research of signature behavior verification algorithmThe handwritten signature is a widely accepted user biometric for authenticating individuals. In order to protect user privacy, this paper proposes a dynamic signature behavior verification method based on touch screen of mobile phone, in which both global and regional features are extracted to identify signatures. In this work, global features mainly consider the relationship between velocity and acceleration in both x,y directions, and the relationship between three consecutive angular from the whole signature. For regional features, basic attributes of a signature are firstly divided into several segments equally. Then, a set of histograms are derived from each segment,including the histograms of velocity, acceleration, and path-tangent angle. Finally, a feature vector can be acquired by combining global features and regional features.The user template is constructed by quantizing multiple user feature vectors with a set of user-specific quantization steps. Then, the dissimilarity score of a test signature to the user template can be computed using Manhattan Distance. The performance of the proposed method is evaluated on the MCYT database and the SG-NOTE database. The experimental results are more accurate than some existing methods.