Research On Visual Information Integrated Visible Light Indoor Localization Algorithms

With the rapid popularization of the mobile smart terminals and the development of wireless communication,navigation and location based services are playing increasingly important roles.Positioning technology is the key to realizing navigation and location based services.Visible light positioning(VLP)technology is an emerging indoor positioning technology based on light-emitting diodes(LEDs).VLP technology relies on Visible Light Communications(VLC)technology,and it can solve the bottleneck problem of the balance between cost and accuracy faced by the existing indoor technologies.However,due to the limitation of the positioning principle,the existing VLP algorithms still has challenges including the orientation limitation on transmitters and receivers,the insufficiency of coverage rate,and the limitation on reliability.Therefore,it is of great significance to study high-accuracy,high-coverage,and high-reliability visible light positioning algorithms.The research of the doctoral dissertation is supported by the project of National Natural Science Foundation of China "Spatial dimming theorems and techniques in visible light communications"(Project Number:61871047)and the National Natural Science Foundation of China"Research on light and visual information combined high accuracy visible light localization theorem and technologies "(Project Number:61901047).Focusing on the problems faced in VLP technology,based on the light information such as signal strength,the doctoral dissertation innovatively proposes the VLP algorithms that integrate visual information such as position of LEDs to gradually achieve high accuracy,high coverage and high reliablity.The main research work and innovations of the doctoral dissertation are as follows:(1)To solve the problem that the VLP algorithms limit the orientation of the transmitter and the receiver,which further affects the positioning accuracy,based on the single view geometry theory,a VLP algorithm that does not limit the receiver orientation and a VLP algorithm that does not limit the orientation of both the receiver and the transmitter are proposed successively.The two algorithms integrate visual information to realizes high-accuracy positioning regardless of the orientation of the receiver and the transmitter.First,to solve the problem that the VLP algorithms limit the orientation of the receiver,a Camera-assisted Received Signal Strength Ratio(RSSR)algorithm(CA-RSSR)is proposed.CA-RSSR proposes to use the visual information captured by the camera to estimate the incident angles of visible lights,and then jointly use the incidence angles and the received signal strength(RSS)measured by the PD to achieve high-accuracy positioning without limitation on the receiver orientation.Simulation and experimental results indicate that CA-RSSR has no limitation on the receiver orientation.For the conventional VLP algorithm,only 40%of positioning samples can achieve the positioning accuracy within 60 cm.In contrast,for CA-RSSR,80%of positioning samples can achieve the accuracy within 9 cm,which means CA-RSSR can achieve a significant advantage in positioning accuracy.Then,to solve the problem that the VLP algorithms limits the transmitter orientation,the RSS-based Perspective-three-Point(P3P)algorithm(R-P3P)is further proposed based on CA-RSSR.In R-P3P,the visual information captured by the camera is used to solve the multiple solution sets of the distance from the LEDs to the receiver,and then the RSSs measured by the PD is used to estimate the right solution set.In this way,R-P3P can futther mitigate the limitation on the transmitter orientation in conventional VLP algorithms without restricting the orientation of the receiver.At the same time,R-P3P can positioning with less beacons compared with the conventional P3P algorithm.Simulation and experimental results indicate that R-P3P has no limitation on the LED orientation.Compared with the conventional VLP algorithm,R-P3P can achieve the coverage gain of more than 20%when the receiver's field of view(FoV)is less than 60 degree.When the LED tilt angle is 30 degree,for CA-RSSR 3D positioning,only 40%of positioning samples can achieve the accuracy within 30 cm.In contrast,for R-P3P,80%of positioning samples can achieve the accuracy within 4 cm.(2)To solve the problem of the insufficient coverage in VLP algorithms,based on the single view geometry theory,the plane geometry theory and the solid geometry theory,a coverage improved VLP algorithm is proposed by integrating visual information.Based on this algorithm,a compensation algorithm is proposed to further improve accuracy,which can achiece high accuracy and high coverage.First,to slove the insufficient coverage problem in VLP algorithms,a Cosine Theorem-assisted RSSR algorithm(CT-RSSR)is proposed.This algorithm deeply uses visual information.Based on the single view geometry theory and the plane geometry theory,the distances between the LEDs and the receiver can be calculated by jointly using light information and visual information.In this way,CT-RSSR can reduce the number of LEDs required for positioning,and reduce the computational complexity by the linear least square method.In other words,CT-RSSR can achieve high-accuracy positioning with high coverage and low complexity.Simulation results indicate that compared with the conventional VLP algorithm,the conventional PnP algorithm and CA-RSSR,CT-RSSR can achieve coverage performance gain of 50%,18%and 42%,respectively.Besides,CT-RSSR can be used in more scenarios than the R-P3P algorithm.Experimental results indicate that CT-RSSR can achieve centimeter-level accuracy with high coverage.Then,to mitigate the effect of the distance between the PD and the camera on positioning accuracy of CA-RSSR and CT-RSSR,an ehanced CT-RSSR(eCT-RSSR)algorithm that compensates the position differences of the receivers is proposed.Based on the single-view geometry theory and the least square algorithm,the receiver position is optimized iteratively.In this way,the positioning error introduced by the distance between the PD and the camera is reduced.Simulation results indicate that eCT-RSSR can reduce the positioning error introduced by the distance between the PD and the camera.When the distance is 20 cm,only less than 60%of positioning samples of CA-RSSR and CT-RSSR can achieve the positioning accuracy within 20 cm.In contrast,80%of positioning samples of eCT-RSSR can achieve the positioning accuracy within 15 cm.(3)To solve the unreliable problem in VLP algorithms,based on the single view geometry theory,the plane geometry theory and the solid geometry theory,a reliability improved VLP algorithm is proposed by integrating visual information.Based on this algorithm,a compensation algorithm is proposed to further improve reliability,which can achieve high accuracy,high coverage and high reliability.First of all,to solve the unreliable problem in VLP algorithms,a VLC-based Perspective-four-Line(P4L)algorithm(V-P4L)is proposed.V-P4L uses a single camera to capture the light information in the time dimension and the visual information in the space dimension at the same time.Based on the geometry theory,the visual information is used to estimate the position and pose of the receiver,which means that V-P4L no longer dependent on the channel model to calculate the distance between the LED and the receiver.Besides,based on the perspective projection theorem,V-P4L can obtain a complete projection of the luminaire under partical occlusion.Therefore,V-P4L also has anti-occlusion ability.In addition,V-P4L use VLC to match the three dimentional(3D)reference features and their two dimensional(2D)projections on the image plane(i.e.,3D-2D correspondences).In this way,V-P4L can solve the problem that vision localization algorithms rely on 3D-2D correspondences.Simulation and experimental results indicate that the high reliability of V-P4L.Compared with CT-RSSR which relies on the channel model and the P4L algorithm which relies on 3D-2D correspondences,V-P4L can achieve the accuracy gain of dozens of centimeters.Then,to solve the problem that V-P4L limits the transmitter orientation,an enhanced V-P4L(eV-P4L)algorithm that compensates the height differences among LEDs is proposed.Based on the single-view geometry theory and the proposed segmentation optimization strategy,eV-P4L can achieve reliable positioning regardless of transmitter orientations.Simulation and experimental results indicate that when the transmitter is tilted,the positioning error of V-P4L gradually increases to more than dozens of centimeters as the tilt angle of the transmitter increases.In contrast,the positioning errors of eV-P4L are always within 20 cm.