Realizing Context-aware Intelligent Mobile Services through Real-Time Indoor Location Tracking

In recent years, the rapid development of mobile technology has led to unprecedented opportunities to realise intelligent environments. The widespread adoption of mobile phone platforms with ever more powerful sensors and processing capabilities has created new opportunities for context aware solutions. This type of environment will enable the user to receive information based on the location and behaviour of the user. Mobile recommender systems, acting as information analysers, can support the decision-making process of users by providing suggestions related to the environment the user is moving through. For example, in a shopping zone, such systems recommend items that are of potential use for customers using both context and behavioural input. Using an in-door real-time tracking system that can track the location of a phone with precision on the centimetre level, allows recommender systems to point users more accurately to shopping items of interest. Further, the high resolution location information can be of help to analyse a user's behaviour, and to offer better learning mechanisms for context-aware mobile shopping. The main objective of the MobiTrack project is to conduct basic research and develop design concepts, software frameworks, and algorithms to utilize real-time position of smartphones in order to improve the intelligence and quality of services that are directly or indirectly location-dependent. The first result achieved in this project describes a product localization algorithm for automated detection of the location of products in a retail store. The proposed learning algorithm uses the data gathered from a user's context to autonomously detect the location of products in the store. The algorithm uses the location at which customers stop when picking up products together with the shopping receipt of the purchased products as inputs. In the second result, DeepMatch was introduced, which is a deep learning model offering highly accurate in-vehicle presence detection using smartphone sensors. This can be used to automatically issue tickets to the passengers based on the exact duration of their ridings on vehicles there is no need for cost intensive ticket checkpoints, ticket machines, and passenger controls anymore. Moreover, it becomes easier to determine the exact flow of passengers between particular places and plan optimal public transport. The third result of MobiTrack, DeepMatch2, is a comprehensive deep learning-based approach for in-vehicle presence detection, in which multiple enhancements to the second result was made. Moreover, this part of the results contains a thorough discussion on travelling user inference systems. The proposed solution is capable of inferring if and for which period of time a passenger makes a trip in a public transport vehicle with a very low error rate. However, an important disadvantage of this approach is that it requires extra hardware, e.g., reference devices and BLE transmitters as fixed equipment installed in busses, which imposes additional costs associated with implementation and maintenance. To address this concern, we, proposed Ataraxis as the fourth result of MobiTrack, a deep learning approach for hardwareless in-vehicle presence detection.

MobiTrack will open new opportunities on how to apply machine learning to obtain context information that was not previously possible. This is thanks to its approach to data analysis in context-aware mobile systems, e.g. for location-aware systems. It will also enable better utilization of sensors on smartphones for analysing behavioural aspects of humans. Moreover, we foresee a significant impact in industry because of automated product localization or in-vehicle presence detection. Regarding the latter, this can revolutionize the way the public transport systems are designed. Hardware-less in vehicle detection is another great achievement of MobiTrack that can significantly reduce the cost of ticketing and optimizing route planning (which will contribute to reducing fuel consumption). The society will also benefit from MobiTrack's results by more incentive using of public transport vehicles and collecting points for the time a passenger has spent in a public transport vehicle, etc.

The widespread adoption of mobile phone platforms with ever more powerful sensors and processing capabilities has created new opportunities for context aware solutions. This type of environment will enable the user to receive information based on the location and behaviour of the user. Mobile recommender systems, acting as information analyzers, can support the decision making process of users by providing suggestions related to the environment the user is moving through. A key source of input information for recommender systems is the real-time location of the user (i.e., the real-time location of user's smartphone). Adoption of this approach has until recently been hampered by the lack of an accurate indoor positioning system for smartphones. With an in-door real-time tracking system that can track the location of a phone with precision on centimeter level, recommender systems can use this high resolution location information to provide more accurate interest of the user to shopping items, analyze his/her shopping behavior, and offer better learning mechanisms for context-aware mobile shopping. Recently Sonitor has developed a solution that will realise this much coveted level of accuracy using a low cost infrastructure and full backwards compatibility with billions of smartphone devices running iOS or Android. In this project, we aim at exploiting this unique capability of Sonitor Ultimate Sense in order to provide more intelligent context-aware recommender systems and consequently make the end-user more satisfied with his/her, e.g., shopping experience, using these systems. This can also be used in predictive modelling (machine learning) to optimize product and advertisement placement in stores.