Urban public space perception behavior analysis method based on Beidou

Abstract

The invention discloses a Beidou-based urban public space perception behavior analysis method, and relates to the technical field of Beidou analysis, and the method comprises the following implementation steps: setting a multi-camera cooperative indoor visual perception space reference model, and obtaining a human body behavior three-dimensional coordinate system; constructing a model according to visual positioning of a human body behavior three-dimensional coordinate system and a pedestrian space trajectory, predicting the probability that each bounding box contains an object by using logistic regression, and selecting an optimal anchor point box to be allocated to a detection target to realize urban public space crowd motion perception recognition; iNS and Beidou navigation are combined, an INS error and a Beidou error are calculated in real time, and the two errors are combined, so that the accuracy of an urban public space perception behavior analysis result based on Beidou is ensured in real time. According to the Beidou-based urban public space perception behavior analysis method, the accuracy of human body perception behavior analysis can be improved, and the problem between the dynamic tracking performance and the anti-interference capability can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of Beidou analysis, and particularly relates to a method for analyzing urban public space perception behavior based on Beidou. Background Technique

[0002] With the acceleration of the urbanization process, the perception of the crowd and the analysis of movement behavior in urban public spaces have become important contents of urban planning and management. The perception of the crowd and the analysis of movement behavior can provide effective support for aspects such as traffic guidance, emergency management, and public safety. The Beidou navigation system is a satellite navigation system independently developed by our country, with advantages such as global coverage and high-precision positioning. Combining the Beidou navigation system and vision technology, conducting research on methods for analyzing the perception of the crowd and movement behavior in urban public spaces has important theoretical and practical significance. The position information provided by the Beidou navigation system can be used to monitor and analyze the distribution, flow, etc. of the crowd.

[0003] Currently, combining the domestic and foreign research status, it is found that the existing research on the spatio-temporal trajectories of the crowd mainly uses the two-dimensional screen pixel trajectories of pedestrians, rather than the three-dimensional precise spatio-temporal trajectories in the real space, which is difficult to reflect the movement characteristics of the crowd in the three-dimensional space. At the same time, the existing technology mainly focuses on the analysis of group trajectories in single-camera videos, lacking research on methods for analyzing the spatio-temporal trajectories of the crowd under cross-camera field-of-view conditions. Therefore, the correlation relationship between the multi-camera system and the real three-dimensional space cannot be directly established, resulting in the inability to reconstruct the three-dimensional visual positioning and spatio-temporal trajectories of the crowd, and the problem of large-scale space crowd perception and monitoring cannot be satisfied. At the same time, there is currently a lack of research on methods for analyzing the spatio-temporal trajectories of the crowd under cross-camera field-of-view conditions, and the abnormal behavior of the crowd cannot be predicted, affecting the efficiency of urban public space perception behavior analysis.

[0004] Therefore, a new method for analyzing urban public space perception behavior based on Beidou is proposed to solve the above problems. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method for analyzing urban public space perception behavior based on Beidou to solve the problems raised in the above background.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is: a method for analyzing urban public space perception behavior based on Beidou, the method includes the following implementation steps:

[0007] Step 1: Set a multi-camera collaborative indoor visual perception space reference model, and combine the basis of the indoor visual perception space reference to construct a coordinate conversion model from pixel coordinates to the real-world coordinate system, calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and obtain the three-dimensional coordinate system of human behavior;

[0008] Step 2: Construct a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the pedestrian spatial trajectory. Use logistic regression to predict the probability that each bounding box contains an object, select the best anchor box to assign to the detection target, and use binary cross-entropy and logistic regression to predict the probability that the target is a pedestrian, so as to achieve the positioning and trajectory tracking of multiple pedestrian targets;

[0009] Step 3: Establish an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network to achieve the recognition of the motion perception of the crowd in the urban public space;

[0010] Step 4: Combine INS and Beidou navigation, and calculate the INS error and Beidou error in real time. Combine the two errors to ensure the accuracy of the analysis results of the Beidou-based urban public space perception behavior in real time.

[0011] The set multi-camera collaborative indoor visual perception space benchmark model constructs a coordinate conversion model from pixel coordinates to the real-world coordinate system in combination with the basis of the indoor visual perception space benchmark. The coordinate conversion model includes four coordinate systems, namely the Beidou positioning global coordinate system, the camera coordinate system, the image coordinate system, and the pixel coordinate system. The Beidou positioning global coordinate system is the world coordinate system, which is expressed by the orthogonal unit rotation matrix R and the three-dimensional translation vector T. It is assumed that the coordinates of a point P in the camera coordinate system and the world coordinate system are (x c , y c , z c ) and (x w , y w , z w ), respectively. Then the conversion relationship is shown in the following formula:

[0012]

[0013] Combining the plane coordinate system and the camera coordinate system, the conversion relationship from the pixel plane coordinate system to the world coordinate system is deduced as follows:

[0014]

[0015] Among them, f, dx, dy, u0, and v0 are the internal parameters of the camera, R and T are the external parameters of the camera, and there is a distortion error in the actual imaging process of camera imaging.

[0016] Combining the influence of distortion, a camera imaging distortion model is constructed. The camera imaging distortion model is expressed by the following formula:

[0017]

[0018] Among them, (u′, v′) are the undistorted ideal image coordinates; (u, v) are the actual image coordinates, k1 and k2 are distortion parameters, and the distortion coefficients are solved by the least squares method. The two-dimensional reprojection error between the actual image coordinates and the calculated image coordinates is shown in the following formula:

[0019]

[0020] Among them, P ij is the image projection point of the corner point j on the i-th image under the camera matrix.

[0021] Use the camera calibration method to accurately calibrate the internal and external parameters of the multi-camera system in the indoor public space, calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and introduce the Beidou ground elevation parameter H into the coordinate transformation model to establish the spatial coordinate calculation model of the pedestrian positioning result as follows:

[0022]

[0023] Among them, z w =H, (u, v) are the pixel point coordinates of the pedestrian's foot bottom. Combine x w and y w to obtain the three-dimensional coordinate system of human behavior (x w , y w , zw).

[0024] The probability that the spatial target is a pedestrian is predicted using binary cross-entropy and logistic regression. The specific method is as follows:

[0025] Assume that the cell target probability is Pr(Object), and determine whether there is a target falling into the candidate box cell. The accuracy can be measured by the overlap degree IOU between the prediction box and the labeled box. is the ratio of the intersection area to the union area of the prediction box and the labeled box, and the value range is between [0,1]. The coordinate error calculation formula between the prediction box and the target labeled box is as follows:

[0026]

[0027] The IOU error is:

[0028] The category error is:

[0029] Among them, x i , y i , w i , h i , C i and p i (c) are the data annotation values. and is the network prediction value, and the loss function loss is obtained through the coordinate error, IOU error, and category error:

[0030]

[0031] Based on the loss function loss, it is assumed that n pedestrians P k，1 , P k，2 ,......, P k，n are detected in the k-th frame, and the pedestrian detection result P k+1，1 , P K+1，2 ,......, P k+1，n is detected in the next frame. The Euclidean distance between the coordinates of pedestrian P k,i and P k+l,j is C ij . The matrix C composed of C ij is called the distance cost matrix. The rows represent the pedestrian coordinates in the k-th frame, and the columns represent the pedestrian coordinates in the K+1-th frame. If pedestrian P k,i matches P k+l,j , it is represented by M ij =1, and if it does not match, it is represented by M ij =0. If M ij =1, then M ix =0 and M yj =0, where x≠j and y≠i;

[0032] The multi-object pedestrian trajectory assignment problem is converted into finding the assignment scheme with the minimum total matching distance S, that is, obtaining the probability minS that the spatial target is a pedestrian:

[0033]

[0034]

[0035] Based on the calculated min S, the Kalman filter is used to correct, update, and predict the trajectory points. The state value at time k is predicted using the optimal value at time k-1, and the calculation formula is as follows:

[0036]

[0037] Predict the new error from the previous error:

[0038]

[0039] Calculate the Kalman gain:

[0040]

[0041] Correction and update equation:

[0042]

[0043] Update the filtering error variance matrix:

[0044]

[0045] Step 5: Establish a pedestrian identity matching model that fuses vision and spatial position. For pedestrian targets moving across camera fields of view, perform visual feature matching to obtain the probability value min S that the predicted spatial target is a pedestrian, and obtain the correction value of the optimal predicted state value at time k. Establish a pedestrian identity matching model that fuses vision and spatial position, and extract the visual features of pedestrians based on the darknet53 network. For pedestrian targets moving across camera fields of view, perform visual feature matching. Constrained by the spatial adjacency relationship of different camera fields of view in the spatial reference, construct the spatio-temporal reachable domain of pedestrians according to the detection time of pedestrians in different camera fields of view, and establish a pedestrian target matching and confidence evaluation method in combination with the pedestrian visual template to achieve continuous positioning and trajectory reconstruction of pedestrians in the indoor global coordinate system.

[0046] The real-time calculation of the INS error and the Beidou error in Step 4 is specifically as follows:

[0047] Where is the attitude error, δv n is the velocity error, δp is the position (latitude, longitude, altitude) error, ε b is the gyroscope error, the accelerometer error;

[0048] Set the distance error equivalent to the clock error as δt u , and set the distance rate error equivalent to the clock frequency error as δt ru , then the error state δt u , δt ru The differential equation calculation formula is as follows:

[0049]

[0050] Where β tru is the autocorrelation time constant, 1 / β tru is the correlation time of the process. Write the above formula in matrix form, that is, as follows:

[0051]

[0052] Where X G (t) = [δt u δt ru T ,​ W G (t)=[ω tu -ω tru T

[0053] Combining the above error state equation with the Beidou error state equation, the system state equation of the pseudorange and pseudorange rate combination system is obtained as follows:

[0054]

[0055] Based on the system state equation, assuming the position of the INS is (x I y I z I ), and the satellite position determined by the satellite ephemeris is (x T , then the pseudorange ρ s y s z s ) T at the position where the INS is located is obtained. At the same time, assuming the pseudorange measured by the Beidou receiver is denoted as ρ I , then the difference between the pseudoranges of the INS and Beidou and the difference between their pseudorange rates are selected as the observables of the integrated navigation system. The pseudorange difference measurement equation is from the pseudorange ρ G from the INS vehicle to the Beidou satellite S j j can be expressed by the following formula: I

[0056]

[0057] Then the pseudorange measured by the Beidou receiver on the vehicle relative to the satellite S j is:

[0058]

[0058] ρ G =r + δt u + υ ρ ;

[0059] The corresponding pseudorange measurement value ρ I of the INS and the Beidou pseudorange ρ G are respectively expressed by the above formula, then the pseudorange difference measurement equation can be written as:

[0060] δρ = ρ I - ρ G + e1 k-1 δx k-1 + e2 k-1 δy k-1 + e3 k-1 δz k-1 = e1 k δx k + e2 k δy k + e3​k δz k -δt ru k -(υ ρ k -υ ρ k-1 );

[0061] Let \(C\) be the coordinate transformation matrix from the navigation system (n) to the geocentric rectangular coordinate system (e). Then, the position errors \(\delta L\), \(\delta\lambda\), and \(\delta h\) expressed in the navigation system can be transformed to the geocentric coordinate system through \(C\). t e Transform to the geocentric coordinate system:

[0062]

[0063]

[0064] where, \(H\) ρ =[0 4×6 H ρ1 0 4×6 H ρ2 4×17 , Obtain the position error data.

[0065] The calculation of the pseudorange rate of change based on the position error data is as follows:

[0066]

[0067] The position coordinate values given by the INS are expressed as the sum of the true value and the error. Thus, the real-time tracking coordinates in the three-dimensional human body coordinate system are:

[0068]

[0069] where, respectively represent the velocity errors in the geocentric rectangular coordinate system;

[0070] The pseudorange rate of change measured by the Beidou receiver is:

[0071]

[0072] Combining the difference result of the pseudorange rates of INS and Beidou, the pseudorange rate measurement equation can be obtained as follows:

[0073] Combining the pseudorange rates of INS and Beidou and using the filtering algorithm to obtain the navigation and positioning solution for the perception behavior analysis of urban public spaces.

[0074] The present invention has the following beneficial effects:

[0075] ​1. In the present invention, by setting up an indoor visual perception spatial reference model with multi-camera collaboration, a coordinate conversion model from pixel coordinates to the real-world coordinate system is constructed based on the indoor visual perception spatial reference, and the spatial fields of view of each camera are calculated in real time. The three-dimensional coordinate system of human behavior can be calculated in real time, realizing the association between multi-cameras and the real three-dimensional space, enabling the construction and analysis of a model based on the indoor space during the analysis of urban public space perception behavior, and ensuring the accuracy of the positioning of the three-dimensional coordinate system of urban public space perception behavior;

[0076] 2. In the present invention, by constructing a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the pedestrian spatial trajectory, and using logistic regression to predict the probability of an object contained in each bounding box, the best anchor box is selected and assigned to the detection template, and the binary cross-entropy and logistic regression are used to predict the probability that the target is a pedestrian, realizing the positioning and trajectory tracking of multiple pedestrian templates, which can meet the problems of large-scale space crowd perception and monitoring, reconstructing the three-dimensional visual positioning and spatio-temporal trajectory of the crowd, and helping to improve the accuracy of human perception behavior analysis;

[0077] 3. In the present invention, by establishing an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network, realizing the recognition of urban public space crowd motion perception, and combining INS and Beidou navigation, calculating the INS error and Beidou error in real time, and combining the two error results, the accuracy of the analysis result of urban public space perception behavior based on Beidou can be ensured in real time. The information fed back by combining INS and Beidou navigation contains the dynamic information of the carrier, which can reduce the dynamics of the carrier tracked by the code loop and carrier loop of the Beidou receiver, thereby increasing the equivalent bandwidth of the code loop and carrier loop, effectively reducing the actual bandwidth of the loop filter, achieving the purpose of suppressing thermal noise, further improving the anti-interference ability of the system in a high-dynamic environment, and effectively solving the problem between dynamic tracking performance and anti-interference ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 is the overall flowchart of a method for analyzing urban public space perception behavior based on Beidou according to the present invention;

[0079] Figure 2 is the schematic diagram of image coordinate conversion of a method for analyzing urban public space perception behavior based on Beidou according to the present invention;

[0080] Figure 3 is the schematic diagram of the architecture of integrated navigation of a method for analyzing urban public space perception behavior based on Beidou according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0081] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0082] Example 1

[0083] Please refer to Figures 1 to 2 as shown below: A Beidou-based method for analyzing urban public space perception behavior, the method includes the following implementation steps:

[0084] Step 1: Set up an indoor visual perception space benchmark model for multi-camera collaboration, and combine the basis of the indoor visual perception space benchmark to construct a coordinate transformation model from pixel coordinates to the real-world coordinate system. Calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and obtain the three-dimensional coordinate system of human behavior;

[0085] Step 2: Construct a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the pedestrian spatial trajectory. Use logistic regression to predict the probability that each bounding box contains an object, select the best anchor box to assign to the detection target, and use binary cross-entropy and logistic regression to predict the probability that the target is a pedestrian, so as to achieve the positioning and trajectory tracking of multiple pedestrian targets;

[0086] Step 3: Establish an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network to achieve the recognition of crowd motion perception in urban public spaces;

[0087] Step 4: Combine INS and Beidou navigation, and calculate the INS error and Beidou error in real time. Combine the two errors to ensure the accuracy of the Beidou-based urban public space perception behavior analysis results in real time.

[0088] Set up an indoor visual perception space benchmark model for multi-camera collaboration, and combine the basis of the indoor visual perception space benchmark to construct a coordinate transformation model from pixel coordinates to the real-world coordinate system. The coordinate transformation model includes four coordinate systems, namely the Beidou positioning global coordinate system, the camera coordinate system, the image coordinate system, and the pixel coordinate system. The Beidou positioning global coordinate system is the world coordinate system, which is expressed by the orthogonal unit rotation matrix R and the three-dimensional translation vector T. Set the coordinates of a point P in the camera coordinate system and the world coordinate system as (x c , y c , z c ) and (x w , y w , z w ), then the conversion relationship is shown in the following formula:

[0089]

[0090] Combining the plane coordinate system and the camera coordinate system, the conversion relationship from the pixel plane coordinate system to the world coordinate system is deduced as follows:

[0091]

[0092] Among them, f, dx, dy, u0, and v0 are internal camera parameters, R and T are external camera parameters. There are distortion errors in the camera imaging during the actual imaging process. By calculating the spatial field of view of each camera in real time, the three-dimensional coordinate system of human behavior can be calculated in real time, realizing the association between multiple cameras and the real three-dimensional space.

[0093] Combining the influence of distortion to construct a camera imaging distortion model, the camera imaging distortion model is expressed by the following formula:

[0094]

[0095] Among them, (u′, v′) are the ideal image coordinates without distortion; (u, v) are the actual image coordinates, k1 and k2 are distortion parameters. By using the least squares method to solve the distortion coefficients, the two-dimensional reprojection error between the actual image coordinates and the calculated image coordinates is shown by the following formula:

[0096]

[0097] Among them, P ij is the image projection point of the corner point j on the i-th image under the camera matrix.

[0098] Using the camera calibration method to accurately calibrate the internal and external parameters of the multi-camera system in the indoor public space, calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and introduce the Beidou ground elevation parameter H into the coordinate transformation model to establish the spatial coordinate calculation model of the pedestrian positioning result as follows:

[0099]

[0100] Among them, z w = H, (u, v) are the pixel point coordinates of the pedestrian's foot bottom. Combining x w and y w to obtain the three-dimensional coordinate system of human behavior (x w , y w , z w ), enabling the model construction and analysis based on the indoor space during the urban public space perception behavior analysis, and ensuring the accuracy of the three-dimensional coordinate system positioning of the urban public space perception behavior.

[0101] Embodiment 2

[0102] Please refer to Figure 2 as shown: Based on the basis of Embodiment 1, use binary cross-entropy and logistic regression to predict the probability that the spatial target is a pedestrian. The specific method is as follows:

[0103] Assume that the target probability of the cell is Pr(Object). To determine whether there is a target in the candidate box cell, the accuracy can be measured by the Intersection over Union (IOU) between the predicted box and the labeled box. It is the ratio of the intersection area to the union area of the predicted box and the labeled box, with a value range between [0, 1]. The coordinate error calculation formula between the predicted box and the target labeled box is as follows:

[0104]

[0105] The IOU error is:

[0106] The class error is:

[0107] Among them, x i , y i , w i , h i , C i and p i (c) are the data annotation values, and are the network prediction values. The loss function loss is obtained through the coordinate error, IOU error, and class error:

[0108] Solving the problem of large-scale spatial crowd perception and monitoring, reconstructing the 3D visual positioning and spatio-temporal trajectories of the crowd, helps to improve the accuracy of human perception behavior analysis.

[0109] Based on the loss function loss, assume that n pedestrians P k，1 are detected in the k-th frame, P k，2 , ……, P k，n , and the pedestrian results P k+1，1 , P K+1，2 , ……, P k+1，n are detected in the next frame. The Euclidean distance between the coordinates of pedestrian P k,i and P k+l,j is C ij . The matrix C composed of C ij is called the distance cost matrix. The rows represent the pedestrian coordinates in the k-th frame, and the columns represent the pedestrian coordinates in the (K + 1)-th frame. If pedestrian P k,i matches P k+l,j , it is represented by M ij = 1, and if not, it is represented by M ij = 0. If M ij = 1, then M ix = 0 and M yj = 0, where x ≠ j and y ≠ i;

[0110] Convert the multi-target pedestrian trajectory allocation problem into finding the allocation plan with the minimum total matching distance S, that is, obtain the probability that the spatial target is a pedestrian, min S:

[0111]

[0112]

[0113] Based on the calculated min S, use Kalman filtering to correct, update, and predict the trajectory points. Predict the state value at time k using the optimal value at time k - 1. The calculation formula is as follows:

[0114]

[0115] Predict the new error from the previous error:

[0116]

[0117] Calculate the Kalman gain:

[0118]

[0119] Correction and update equation:

[0120]

[0121] Update the filtering error variance matrix:

[0122]

[0123] Step 5: Establish a pedestrian identity matching model that fuses vision and spatial position. Perform visual feature matching on pedestrian targets moving across camera fields of view to obtain the probability value min S that the predicted spatial target is a pedestrian, and obtain the correction value of the optimal value for predicting the state value at time k Establish a pedestrian identity matching model that fuses vision and spatial position. Based on the darknet53 network, extract the visual features of pedestrians. Perform visual feature matching on pedestrian targets moving across camera fields of view. Using the spatial adjacency relationship of different camera fields of view in the spatial reference as a constraint, construct the spatio-temporal reachable domain of pedestrians according to the detection time of pedestrians in different camera fields of view. Combine the pedestrian visual template to establish a pedestrian target matching and confidence evaluation method to achieve continuous positioning and trajectory reconstruction of pedestrians in the indoor global coordinate system.

[0124] Example 3

[0125] Please refer to Figure 3 As shown: Based on the foundation of Example 1, in Step 4, calculate the INS error and Beidou error in real time. The specific INS error calculation formula is as follows:

[0126]

[0127] Among them, is the attitude error, δv n is the velocity error, δp is the position (latitude, longitude, altitude) error, ε b is the gyroscope error, the accelerometer error;

[0128] Set the distance error equivalent to the clock error as δt u , and set the distance rate error equivalent to the clock frequency error as δt ru , then the error state δt of Beidou u 、δt ru The differential equation calculation formula is as follows:

[0129]

[0130] Among them, β tru is the autocorrelation time constant, 1 / β tru is the correlation time of the process. Write the above formula in matrix form, that is, as follows:

[0131]

[0132] Among them, X G (t) = [δt u δt ru T , W G (t) = [ω tu -ω tru T

[0133] Merge the above error state equation with the Beidou error state equation, then the system state equation of the pseudorange and pseudorange rate combined system is obtained, as follows:

[0134]

[0135] Based on the system state equation, set the position of the INS as (x I y I z I ), T and the satellite position determined by the satellite ephemeris is (x s y s z s ), T then the pseudorange ρ at the position where the INS is located is obtained I . At the same time, set the pseudorange measured by the Beidou receiver as ρ G ​​, the pseudorange difference and the pseudorange rate difference between INS and Beidou are selected as the observables of the integrated navigation system. The pseudorange difference measurement equation is from the INS vehicle to the Beidou satellite S j The pseudorange ρ Ij Can be expressed by the following formula:

[0136]

[0137] Then the pseudorange measured by the Beidou receiver on the vehicle relative to the satellite S j Is:

[0138] ρ G = r + δt u + υ ρ ;

[0139] The corresponding pseudorange measurement value ρ of INS I And the Beidou pseudorange ρ G Are respectively expressed by the above formula, then the pseudorange difference measurement equation can be written as:

[0140] δρ = ρ I - ρ G + e1 k-1 δx k-1 + e2 k-1 δy k-1 + e3 k-1 δz k-1 = e1 k δx k + e2 k δy k + e3 k δz k - δt ru k -(υ ρ k - υ ρ k-1 );

[0141] Let be the coordinate transformation matrix from the navigation system (n) to the geocentric rectangular coordinate system (e). Then the position errors δL, δλ, δh expressed in the navigation system can be transformed to the geocentric coordinate system through Transformation:

[0142]

[0143]

[0144] Where, H ρ = [0 4×6 H ρ1 0 4×6 H ρ2 4×17 , ​The position error data is obtained, and the INS error and Beidou error are calculated in real time. By combining the two error results, the accuracy of the urban public space perception behavior analysis based on Beidou can be guaranteed in real time. The information fed back by combining INS and Beidou navigation contains the dynamic information of the carrier.

[0145] Calculate the rate of change of pseudorange according to the position error data. The calculation formula of the rate of change of pseudorange is as follows:

[0146]

[0147] The position coordinate value given by INS, expressed as the sum of the true value and the error, so there is the real-time tracking coordinate in the three-dimensional coordinate system of the human body:

[0148]

[0149] Among them, respectively represent the velocity errors in the geocentric rectangular coordinate system;

[0150] The rate of change of pseudorange measured by the Beidou receiver is:

[0151]

[0152] Combining the difference result of the pseudorange rates of INS and Beidou, the pseudorange rate measurement equation can be obtained as follows:

[0153] Combining the pseudorange rates of INS and Beidou, and using the filtering algorithm to obtain the navigation and positioning solution for the urban public space perception behavior analysis. The information fed back by combining INS and Beidou navigation contains the dynamic information of the carrier, which can reduce the dynamics of the carrier tracked by the code loop and carrier loop of the Beidou receiver, thereby increasing the equivalent bandwidth of the code loop and carrier loop, effectively reducing the actual bandwidth of the loop filter, achieving the purpose of suppressing thermal noise, and further improving the anti-interference ability of the system in a high-dynamic environment.

[0154] In the present invention, a method for analyzing urban public space perception behavior based on Beidou first sets a multi-camera collaborative indoor visual perception space benchmark model, constructs a coordinate transformation model from pixel coordinates to the real-world coordinate system based on the indoor visual perception space benchmark, calculates the spatial field of view of each camera, establishes a spatial mapping relationship between the camera field of view and the indoor global coordinate system, obtains the three-dimensional coordinate system of human behavior. By setting the multi-camera collaborative indoor visual perception space benchmark model and constructing the coordinate transformation model from pixel coordinates to the real-world coordinate system based on the indoor visual perception space benchmark, and calculating the spatial field of view of each camera in real time, the three-dimensional coordinate system of human behavior can be calculated in real time, realizing the association between multiple cameras and the real three-dimensional space, enabling the construction and analysis of models based on the indoor space during the analysis of urban public space perception behavior, and ensuring the accuracy of the three-dimensional coordinate system positioning of urban public space perception behavior; constructs a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the pedestrian spatial trajectory, uses logistic regression to predict the probability that each bounding box contains an object, selects the best anchor box to assign to the detection target, and uses binary cross-entropy and logistic regression to predict the probability that the target is a pedestrian, realizing the positioning and trajectory tracking of multiple pedestrian targets. By constructing a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the pedestrian spatial trajectory, using logistic regression to predict the probability that each bounding box contains an object, selecting the best anchor box to assign to the detection template, and using binary cross-entropy and logistic regression to predict the probability that the target is a pedestrian, the positioning and trajectory tracking of multiple pedestrian templates can be realized, which can meet the problems of large-space crowd perception and monitoring, reconstruct the three-dimensional visual positioning and spatio-temporal trajectory of the crowd, and help improve the accuracy of human perception behavior analysis; establishes an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network to realize the recognition of urban public space crowd motion perception; combines INS and Beidou navigation, and calculates the INS error and Beidou error in real time, combines the two errors, and ensures the accuracy of the analysis results of urban public space perception behavior based on Beidou in real time. By establishing an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network to realize the recognition of urban public space crowd motion perception, combining INS and Beidou navigation, calculating the INS error and Beidou error in real time, and combining the two error results, the accuracy of the analysis results of urban public space perception behavior based on Beidou can be ensured in real time. The information fed back by combining INS and Beidou navigation contains the dynamic information of the carrier, which can reduce the dynamics of the carrier tracked by the code loop and carrier loop of the Beidou receiver, thereby increasing the equivalent bandwidth of the code loop and carrier loop, effectively reducing the actual bandwidth of the loop filter, achieving the purpose of suppressing thermal noise, further improving the anti-interference ability of the system in a high-dynamic environment, and effectively solving the problem between dynamic tracking performance and anti-interference ability.

[0155] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention as claimed. The scope of the present invention as claimed is defined by the appended claims and their equivalents.

Claims

1. A Beidou-based urban public space perception behavior analysis method, characterized in that: The method comprises the following implementation steps: Step 1: Set up a multi-camera collaborative indoor visual perception space benchmark model, and build a coordinate transformation model from pixel coordinates to the real-world coordinate system based on the indoor visual perception space benchmark, calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and obtain the three-dimensional coordinate system of human behavior; Step 2: Build a model based on the visual positioning of the three-dimensional coordinate system of human behavior and the spatial trajectory of pedestrians, use logistic regression to predict the probability that each bounding box contains an object, select the best anchor box to assign to the detection target, use binary cross entropy and logistic regression to predict the probability that the target is a pedestrian, and achieve positioning and trajectory tracking of multiple pedestrian targets; Step 3: Establish an indoor group motion behavior classification model based on the long short-term memory network in the recurrent neural network to realize the perception and recognition of crowd motion in urban public spaces; Step 4: Combine INS and Beidou navigation, calculate INS error and Beidou error in real time, combine the two errors, and ensure the accuracy of Beidou-based urban public space perception behavior analysis results in real time.

2. The method according to claim 1, characterized in that: The method sets a multi-camera collaborative indoor visual perception space benchmark model, and combines the indoor visual perception space benchmark to build a coordinate transformation model from pixel coordinates to real-world coordinate systems. The coordinate transformation model includes four coordinate systems, namely, the Beidou positioning global coordinate system, the camera coordinate system, the image coordinate system, and the pixel coordinate system. The Beidou positioning global coordinate system is the world coordinate system, which is expressed by an orthogonal unit rotation matrix R and a three-dimensional translation vector T. The coordinates of a point P in the space in the camera coordinate system and the world coordinate system are respectively (x c ,y c , z c ) and (x w ,y w , z w ), the conversion relationship is as follows: Combining the plane coordinate system and the camera coordinate system, the conversion relationship from the pixel plane coordinate system to the world coordinate system is derived as follows: Among them, f, dx, dy, u0 and v0 are the intrinsic parameters of the camera, R and T are the extrinsic parameters of the camera, and there is distortion error in the actual imaging process of the camera.

3. The method according to claim 2, characterized in that: The camera imaging distortion model is constructed by combining the distortion effect. The camera imaging distortion model is expressed by the following formula: Among them, (u′, v′) is the ideal image coordinate without distortion; (u, v) is the actual image coordinate, k1 and k2 are distortion parameters, and the distortion coefficient is solved by the least squares method. The two-dimensional reprojection error between the actual image coordinate and the calculated image coordinate is shown as follows: Among them, P ij is the image projection point of corner point j on the i-th image under the camera matrix.

4. The method according to claim 3, characterized in that: The camera calibration method is used to accurately calibrate the internal and external parameters of the multi-camera system in the indoor public space, calculate the spatial field of view of each camera, establish the spatial mapping relationship between the camera field of view and the indoor global coordinate system, and introduce the Beidou ground elevation parameter H into the coordinate transformation model to establish the spatial coordinate calculation model of the pedestrian positioning result as follows: Among them, z w =H, (u, v) is the pixel coordinate of the pedestrian's foot bottom, and x w and w The three-dimensional coordinate system of human behavior (x w ,y w , z w ).

5. The method according to claim 1, characterized in that: The method of using binary cross entropy and logistic regression to predict the probability that the spatial target is a pedestrian is as follows: Assuming that the cell target probability is Pr(Object), determine whether there is an object falling into the candidate box cell. The accuracy can be measured by the overlap IOU between the prediction box and the marked box. It is the ratio of the intersection area of ​​the prediction box and the marked box to the union area, and its value range is between [0,1]. The coordinate error calculation formula between the prediction box and the target marked box is as follows: The IOU error is: The category error is: Among them, x i ,y i 、w i 、h i , C i and p i (c) is the data annotation value, and For the network prediction value, the loss function loss is obtained through coordinate error, IOU error and category error: Based on the loss function loss, assume that n pedestrians P are detected in the kth frame. k，1 , P k，2 ,......,P k，n , the pedestrian result P is detected in the next frame k+1，1 , P k+1，2 ,......,P k+1，n , pedestrian P k,i With P k+l,j The Euclidean distance of the coordinates is C ij , by C ij The matrix C is called the distance cost matrix. The rows represent the coordinates of the pedestrians in the kth frame, and the columns represent the coordinates of the pedestrians in the K+1th frame. k,i With P k+l,j If it matches, use M ij =1, if it does not match, use M ij =0 means, if M ij =1, then M ix =0 and M yj =0, where x≠j, y≠i; The multi-target pedestrian trajectory allocation problem is converted into finding an allocation scheme with the minimum matching total distance S, that is, the probability min S that the spatial target is a pedestrian is obtained:

6. The method according to claim 5, characterized in that: Based on the calculated min S, the Kalman filter is used to correct, update and predict the trajectory points, and the optimal value at time k-1 is used to predict the state value at time k. The calculation formula is as follows: Predict the new error from the previous error: Calculate the Kalman gain: Correction update equation: Update the filter error variance matrix: Step 5: Establish a pedestrian identity matching model that integrates vision and spatial position, perform visual feature matching on pedestrian targets that move across the camera field of view, obtain the probability value min S that the predicted spatial target is a pedestrian, and obtain the correction value of the state value at the optimal value prediction k moment A pedestrian identity matching model that integrates vision and spatial position is established, and the visual features of pedestrians are extracted based on the darknet53 network. Visual feature matching is performed on pedestrian targets that move across camera fields of view. The spatial adjacency relationship between different camera fields of view in the spatial benchmark is used as a constraint. According to the detection time of pedestrians in different camera fields of view, the spatiotemporal reachable domain of pedestrians is constructed. A pedestrian target matching and confidence evaluation method is established in combination with pedestrian visual templates to achieve continuous positioning and trajectory reconstruction of pedestrians in the indoor global coordinate system.

7. The method according to claim 1, characterized in that: In the step 4, the INS error and Beidou error are calculated in real time. The specific INS error calculation formula is as follows: in, is the attitude error, δv n is the velocity error, δp is the position (latitude, longitude, altitude) error, ε b is the gyroscope error, Accelerometer error; Assume that the distance error equivalent to the clock error is δt u , set the distance rate error equivalent to the clock frequency error to be δt ru , then the error state of BeiDou is δt u ,δt ru The differential equation for is as follows: Among them, β tru is the autocorrelation time constant, 1 / β tru is the relevant time of the process, and the above formula is written in the form of a matrix, which is as follows: Among them, X G (t)=[δt u δt ru ] T , W G (t) = [ω tu -ω tru ] T Combining the above error state equation with the Beidou error state equation, the system state equation of the pseudorange and pseudorange rate combined system is obtained as follows:

8. The method according to claim 7, characterized in that: Based on the system state equation, the position of INS is assumed to be (x I y I z I ) T , the satellite position determined by the satellite ephemeris is (x s y s z s ) T , then the pseudorange ρ at the location of INS is obtained I At the same time, let the pseudorange measured by the Beidou receiver be denoted as ρ G , then the difference between the pseudoranges of INS and BeiDou and the difference between their pseudorange rates are selected as the observation quantities of the integrated navigation system, where the pseudorange difference measurement equation is from the INS carrier to the BeiDou satellite S j The pseudorange ρ Ij It can be expressed by the following formula: Then the Beidou receiver on the carrier is relative to the satellite S j The measured pseudorange is: r G =r+δt u +u ρ ; INS corresponding pseudorange measurement value ρ I Pseudorange with Beidou G Expressed by the above formulas respectively, the pseudo-range difference measurement equation can be written as: δρ=ρ I -r G +e1 k-1 δx k-1 +e2 k-1 dy k-1 +e3 k-1 δz k-1 =e1 k δx k +e2 k dy k +e3 k δz k -δt ru k -(u ρ k -u ρ k-1 ); Assume that the coordinate transformation matrix from the navigation system (n) to the geocentric rectangular coordinate system (e), then the position errors δL, δλ, δh expressed in the navigation system can be expressed by Transform to geocentric coordinates: Among them, H ρ =[0 4×6 H ρ1 0 4×6 H ρ2 ] 4×17 , Get the position error data.

9. The method according to claim 8, characterized in that: The pseudorange change rate is calculated based on the position error data, and the calculation formula of the pseudorange change rate is as follows: The position coordinate value given by INS is expressed as the sum of the true value and the error, so there is the real-time tracking coordinate of the human body's three-dimensional coordinate system: in, They represent the velocity errors in the geocentric rectangular coordinate system respectively; The pseudorange change rate measured by the BeiDou receiver is: Combining the pseudo-range rate difference results of INS and Beidou, the pseudo-range rate measurement equation can be obtained as follows: By combining the pseudo-range rate of INS and BeiDou, the navigation and positioning solution for urban public space perception behavior analysis is obtained.

Images