Indoor/Outdoor Detection By Wearable Device In Environmental Sensing

Context-aware computing is the key technologies for intelligent interaction applications,and is also an important subject in the field of Internet of Things.In real life,service provided by smart devices is different from situations to situations.And context awareness provides basic environmental information for the upper application,which enables the application to adjust the running strategy according to the environment information,which makes the service more intelligent.With the development of mobile Internet,the context of many applications is more diversified and dynamic.However,the way of data collection stays unchangeable,making the collected data contains different context information.That brings inconvenience for application of data in different situations.And the increased amount of data poses a great challenge for data processing.In this paper,we study the sensing data collected by the weather wearable devices under the condition of existing weather data collection and try to study the indoor/outdoor context detection for sensor data.Most existing methods for indoor/outdoor detection need to manually label data before detecting.However,it is not practical to manually label the data during the large scale collection.And it is very challenging for supervised or semi-supervised methods.In this paper,we present an unsupervised method for indoor/outdoor(IO)detection.The method builds gradient time series model on multidimensional sensor data collected by the wearable devices.The context switching patterns are analyzed for capturing the indoor/outdoor context switching points.According the points,the sensor data time series are divided into several subsequences.And each subsequence contains one local context.And then,a similarity model is built to calculate the similarity between each subsequence and reference time series with outdoor context.Then,the context of subsequence is decided by the calculated similarity.If the similarity is small enough,the subsequence contains the same context with reference time series.We validate our method in a real environment with our sensors.The result shows that the classification accuracy of our method is high in large scale data collection.The contribution of this paper are showed as following:In this paper,we propose an unsupervised method for indoor/outdoor detection via wearable devices equipped with multiple sensors.Compared with exiting supervised method,the method proposed in this paper needs not manually label data and is suitable for large data collection situation.In addition,the proposed method is modeled time series on sensor data and constructs a switching context points detection algorithm.According to the captured switching context points,the sensor data time series are divided into several subsequences.Then,the similarity between sensor data subsequence and reference time series is calculated for indoor/outdoor detection.Compared to exiting indoor/outdoor detection based on rules,the method is more adaptive in different situations.