Research On Indoor Localization Using Sparselyly Networked Cameras And Smart Phones

Location-Based Service (LBS) is an important and widely concerned application related to the pervasive computing technique, which has been deeply integrated into our lives. In order to put LBS into better real-use, accurate determining the locations of people (localization or positioning of people) needs to be completed first. In comparison with the outdoor environment, the complex indoor environment poses more challenges on exploring indoor localization methods. Camera networks and mobile phones are widely utilized for smart homes, intelligent hospitals, etc., which would relieve the burden of the doctors and nurses. However, the cameras are usually deployed very sparsely for covering a large area, leading to the cameras have gaps between each other, and provide less accurate results in the gap in comparison with in the FoVs because of the lack of measurements. Meantime, smart phones are becoming popular devices for seamless indoor localization. Yet, the disadvantages are the inaccurate sensors, which decrease the accuracy of the localization results. Consequentely, joining sparsely networked cameras and smart phones for indoor localization would combine the advantages of each one, get rid of the disadvantages, and demonstrate superior performance. And therefore, aiming at monitoring the people who need special care in hospitals or places where they live, an accurate indoor localization method by making use of the widely utilized equipments in such scenarios, sparsely networked cameras and mobile phones carried by the interested targets, will be considered in this dissertation. And the main contents of this dissertation can be summarized as follows:An effective mechanism is proposed to address the pratical problems of the Simultaneous Localization And Tracking (SLAT) using maximum a posteriori estimation method for determining the sparsely networked cameras’ extrinsic parameters (i.e. the position and orientation of the cameras). Among the popular approaches, the SLAT method has been shown to work effectively with less constraints compared to alternative approaches. However, poor robustness of the method inspires us to develop a mechanism to address these problems. The simulation results demonstrate the better calibration results can be obtained with the mechanism than without it.This thesis proposes Location Constrained Extrinsic Calibration (LCEC) method for disjoint networked cameras using maximum a posteriori estimation. First, several points with known locations are randomly deployed in the area covered by the cameras. Second, an assisted calibration object is controlled to freely move in the surveillance area and opportunistically pass the field of views of the cameras and some of the points. Finally, we formulate the localization problem as a joint inference of the extrinsic parameters and object trajectory based on the cameras’ observations and the points the object passed by. Results herein demonstrate the superior performance of the proposed method over the state-of-the-art methods based on the MABR and classical metric in simulations and real experiments.Intended for identifying multiple people, Accelerometers and Gyroscopes in a Smart Phone assisted multiple People Identification method (AGSPPI) is developed by making use of accelerometers and gyroscopes embedded in smart phones and a camera. The proposed multiple people identification method first computes walking speeds and heading directions for the captured objects based on the camera’s observations. Then, walking speeds and heading directions corresponding to the objects in the surveillance area are deduced in another approach by making use of the measurements from their own smart phones. Since each smart phone has a unique identity (International Mobile Equipment Identity，IMEI), the walking speeds and heading directions derived from smart phone are labeled with the corresponding identities. Therefore, the successful identification of different people captured by the camera is achieved by correctly matching the walking speeds and heading directions obtained with the two different meanings. The results obtained from the simulations and real experiments verify the good performance of this method.Using smart phones and sparsely networked cameras, a robust indoor localization method Multiple People Identification and Localization using Smart phones and Sparsely Networked Cameras (MPILSSNC) is explored. The accelerometers, gyroscopes, magnetometers and WiFi receiver embedded in a smart phone are chosen to be cooperated with sparsely networked cameras for developing the indoor localization method. Intended for obtaining accurate localization results, we first improve the Pedestrian Dead Reckoning (PDR) technique based on the accelerometers, gyroscopes and magnetometers. And then the accumulative error of the PDR technique is decreased by opportunistically resetting the locations of the interested people in the following two approaches:(1) Given cameras’ observations, the captured people’s locations will be calculated by first determining the possible identities with the aid of Receive Signal Strength (RSS) and then simultaneous identifying and localizing them using the proposed multiple people identification method.(2) Given the sink node receives RSS values from smart phones and the sink node does not receive the corresponding cameras’ measurements, the locations of the smart phones will be computed based on the RSS and PDR technique using the Markov Chain Monte Carlo (MCMC) technique. The real experiments’ results demonstrate the superior performance of the proposed method.