Research On Visual Indoor Localization Algorithm Based On Spatial-Temporal Constraints

With the rapid development of mobile Internet,accurate indoor localization is in great need and has a board range of applications,such as public security,medical services and offline commerce.At present,most of the indoor localization technologies are based on wireless RF signals such Wi-Fi and Bluetooth.However,measured RF signals often suffer from multipath effects,which may degrade the localization accuracy significantly.Since visual landmarks are relatively stable over a long period of time and do not require additional infrastructure,vision-based indoor localization has received much attention recently.At present,existing vision-based indoor localization technologies usually require users to take multiple independent images around his/her location and manually confirm the shortlisted landmarks.The process is inconvenient,slow,and unnatural.Moreover,due to the varied building structures,dynamic indoor environment,the accuracy,robustness and usability of current indoor localization techniques are compromised,hindering the development of localization services.To address the above,we propose a visual indoor localization algorithm based on spatial-temporal constraints,which is user-friendly,efficient and high-precision.To solve the problem of usability,the algorithm utilizes object detection and tracking techniques and selects key frames of video automatically and efficiently based on the temporal consistency constraints.To reduce the influence of scene lighting changes and motion blur on the landmark identification,the algorithm uses the prior information provided by indoor map to formulate Hidden Markov Model and filters the misidentified landmarks based on the spatial consistency constraints.Suffered from the noisy compass caused by complex indoor environment,the proposed algorithm takes the compass noise into consideration and constructs an optimization model based on triangulation,which improves the robustness of localization system.In addition,we design and develop a vision-based localization system based on the proposed algorithm.The effectiveness of the algorithm is verified by extensive experiments in two indoor sites,which our proposed scheme is fast(less than 1 second to localize client)and achieves substantially better accuracy as compared with thestates-of-the-art(reducing the localization error by 30%).