Active-passive Cooperative Indoor Location Method Combining Visual And Inertial Features

High-accuracy indoor positioning is a general research subject in many fields.In this paper,indoor positioning is divided into active and passive positioning in the aspect of implementation mechanism.For the advantages and disadvantages of active and passive positioning,this paper intends to select smart phone inertial data and ordinary monocular surveillance video data as active and passive positioning information sources respectively,which can reach to a new method of co-location in sub-meter positioning accuracy without high-cost equipment.With the advantage of visual positioning-high accuracy,the new method of active-passive co-location proposed in this paper is mainly based on visual positioning and supplemented by inertial positioning.Visual positioning is based on pedestrian detection,but the result of pedestrian detection is pixel coordinates,rather than world coordinate system(which is used to describe the real space position information of the pedestrian).For this issue,this paper analyzes the camera imaging principle and camera calibration method,and designs a scene-assisted monocular camera coordinate transformation model to realize the conversion of "pixel-world" coordinates.Experiments show that the conversion model can achieve centimeter-level accuracy.In the real-time processing of visual positioning,each frame of image data will get the pixel positioning result of multiple pedestrians.Therefore,it is necessary to complete the distinction of multi-personal results and to achieve continuous tracking of multiple single pedestrians in the data of consecutive frames.In this paper,combing with the Hungarian algorithm and Kalman filter,a multi-row continuous tracking system based on pedestrian detection technology is designed.Based on the "pixel-world" coordinate transformation model,the continuous tracking and positioning of multiple pedestrians is completed.The tracking system proved by experiments is effective and has strong robustness.Due to the lack of visual positioning(visual blind spot,unknown pedestrian identity,abnormal positioning,and so on.),it is necessary to use inertial information to effectively supplement.Inertial positioning relies on pedestrian trajectory estimation.This paper has done research on step size estimation,step detection and heading angle calculation.During the movement of pedestrians,there will be a unique sequence of motion features,which will gradually enrich the information over time.In view of this,this paper proposes the concept of pedestrian motion time series feature,and describes the motion time series features using two dimensions(inertial and visual)data,designing the matching algorithm of two feature sequences,and accomplishing the unification of the two kinds of information through sequence matching.Based on the unified information foundation,this paper designs an active-passive co-location system combining visual and inertial features.Experiments show that the co-location system can effectively utilize the high-precision advantages of visual positioning and the short-time high-precision characteristics of inertial positioning to achieve indoor multi-personal sub-meter positioning accuracy.