A Complex User Activity Recognition Algorithm based on Convolutional Neural Networks

摘要: Using sensors to accurately identify the user’s activity is helpful select suitable indoor localization algorithm, thus improving positioning accuracy. Traditional user recognition technology generally needs artificially analyze and extract data features, sometimes it difficult excavate characteristics of complex activity. By making use ability automatically extracting feature raw with deep learning this paper proposes a novel algorithm for based on CNN. As representative neural network learning, CNN provides an end-to-end model high The utilizes original collected by embedded in smartphone, such as acceleration sensor, gyroscope sensor magnetic input after 2D image-like conversion. training optimizing various super parameters convolution network, layer number filters, features different activities are learned At same time, we ReLU activation function improve speed, finally build multiple network. comparison, implements AdaBoost using handcrafted features. Extensive experimental results show that our proposed can six kinds behavior modes, walking normallywalking sideways, backward, crawling floor, pushing pushcart, going up down elevator, accuracy 99.8%, which superior