A MCMC road roughness detection method based on point cloud

Abstract

The purpose of the present invention is to provide a MCMC road roughness detection method based on point cloud. First, point cloud data is used for sampling to form samples, and the state of the samples is defined and initialized. Then, a Markov chain is constructed and a state transfer matrix is ​​calculated. Each state represents the proportion of different road levels in the sample. The calculation result of the state transfer matrix can be used to simulate and capture the transfer relationship between different levels of road roughness, and obtain a more comprehensive road roughness distribution from a statistical point of view.

Description

Technical Field

[0001] The present invention relates to the technical field of road detection, and in particular to a MCMC road roughness detection method based on point cloud. Background Art

[0002] Monte Carlo method is a method to solve various mathematical problems through random sampling and statistical analysis. Monte Carlo method can be used to estimate numerical values, solve probability problems, simulate system behavior, etc. Monte Carlo method usually includes random sampling, integration, simulation, Markov chain Monte Carlo and other techniques. Markov Chain Monte Carlo method (MCMC) is a special form of Monte Carlo method, which uses Markov chain to generate random samples to estimate the properties of complex probability distribution.

[0003] Monte Carlo methods usually require the generation of a large number of random samples to ensure the accuracy and stability of the estimation results. As the number of samples increases, the estimation results of the Monte Carlo method usually approach the true value, and the estimation error decreases with the increase in the number of samples. For complex problems or integral calculations in high-dimensional space, a large number of samples are usually required to obtain accurate estimates. Compared with the general Monte Carlo method, the MCMC method can usually achieve sampling of the posterior distribution of parameters with a smaller number of samples.

[0004] MCMC methods use the properties of Markov chains to perform random walks in parameter space, thereby effectively exploring parameter space and converging to the target distribution. Even when the number of samples is relatively small, MCMC methods are able to provide reasonable estimates of the posterior distribution of parameters because they utilize the correlation information between parameters.

[0005] MCMC is often used in Bayesian statistical inference, probabilistic graphical models, Monte Carlo integration, etc. The main idea of ​​MCMC is to construct a Markov chain so that its stationary distribution is equal to the probability distribution of interest, and then sample from the chain to finally obtain a sample set to estimate various statistics.

[0006] In general, Monte Carlo methods are a broader set of stochastic simulation methods, while Markov Monte Carlo is a subfield of Monte Carlo methods that focuses on the use of Markov chains to deal with complex probability distribution estimation problems.

[0007] In traditional road roughness detection, the roughness of the road ahead is regarded as a problem of finding the optimal solution. However, when the vehicle is driving, the tires cannot pass all the roads ahead. Therefore, in the present invention, the roughness of the road ahead is regarded as a probability distribution problem. By sampling and analyzing the data point cloud of the road ahead, the PSD of the sample points is statistically analyzed, and the PSD distribution of the roads at each level of the sample points is obtained. The suspension control is performed according to the component proportions of each level of the roads. Summary of the invention

[0008] The purpose of the present invention is to provide a MCMC road roughness detection method based on point cloud, which uses point cloud data for sampling to form samples, defines and initializes the states of the samples, and then constructs a Markov chain and calculates the state transfer matrix. Each state represents the proportion of different road levels in the sample. The calculation result of the state transfer matrix can be used to simulate and capture the transfer relationship between different levels of road roughness, and obtain a more comprehensive road roughness distribution from a statistical point of view.

[0009] The specific implementation steps of the present invention are as follows:

[0010] S1: Random selection of samples;

[0011] S2: Definition of state;

[0012] S3: initialization of state;

[0013] S4: Construction of Markov chain;

[0014] S5: calculation of state transfer matrix;

[0015] S6: Verification of the effectiveness of the method.

[0016] Furthermore, in the above S1, the road point cloud is three-dimensional data, and each of its data points represents the roughness of the road in front of the vehicle; first, it is necessary to sample the point cloud, perform roughness component analysis on the entire road according to the sampling sequence, and use random sampling method to obtain data samples.

[0017] Furthermore, in the above S2, after the data samples are obtained, the data samples are samples in the spatial domain, and their representation needs to be converted into the frequency domain.

[0018] Furthermore, in the above S3, after converting the spatial domain samples into the frequency domain, all the frequency components are classified into road grades, and the proportions of A, B, C, D, E, and F roads are counted.

[0019] Furthermore, in the above S4, after the basic state is obtained, under the condition that the Markov property is verified, the state space formed by the above S1, S2, and S3 is regarded as a Markov process.

[0020] Furthermore, in the above S5, calculating the state transition matrix of any two states in the state space is the core task of the Markov chain model; this matrix describes the probability of transitioning from one state to another, and the transition probability converges to a stable value, indicating that the transition relationship between roads remains stable under the current state, that is, the relative roughness between roads of different levels maintains a certain relationship, which helps to understand the overall condition of the road.

[0021] Furthermore, in the above S6, a dataset of labeled simulated point cloud dataset with 5% Gaussian white noise added is used to verify the effectiveness and accuracy of the algorithm.

[0022] The beneficial effects of the present invention are as follows:

[0023] (1) The MCMC method is introduced into the detection of road roughness. By sampling the road point cloud to form a state space, a Markov chain is constructed, the state transfer matrix is ​​calculated, and the road roughness level distribution of the road point cloud is determined;

[0024] (2) The power spectral density of the sample data is estimated to determine the proportion of each level of road in the sample data. The analysis results of the roughness of the sample data are regarded as a single state, and the Markov property between the states is verified;

[0025] (3) The method is verified using a labeled simulation point cloud dataset. The experimental results show the effectiveness and accuracy of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of the specific steps of the present invention;

[0027] Figure 2 It is a line graph of the verification accuracy of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The purpose of the present invention is to provide a MCMC road roughness detection method based on point cloud, which uses point cloud data for sampling to form samples, defines and initializes the states of the samples, and then constructs a Markov chain and calculates the state transfer matrix. Each state represents the proportion of different road levels in the sample. The calculation result of the state transfer matrix can be used to simulate and capture the transfer relationship between different levels of road roughness, and obtain a more comprehensive road roughness distribution from a statistical point of view.

[0030] As attached Figure 1 As shown, the specific implementation steps of the present invention are as follows:

[0031] S1: Random selection of samples;

[0032] S2: Definition of state;

[0033] S3: initialization of state;

[0034] S4: Construction of Markov chain;

[0035] S5: calculation of state transfer matrix;

[0036] S6: Verification of the effectiveness of the method.

[0037] Furthermore, in the above S1, the road point cloud is three-dimensional data, and each of its data points represents the roughness of the road in front of the vehicle; first, it is necessary to sample the point cloud, perform roughness component analysis on the entire road according to the sampling sequence, and use random sampling method to obtain data samples.

[0038] S101: Each data point in the point cloud is independent of each other and does not affect each other. Included points, and the probability of each point being selected is .

[0039] S102: Extraction from point cloud data points, of which , forming a road roughness sample data , then the sample space , sample sampling rate , which can be expressed as:

[0040]

[0041] Furthermore, in the above S2, after the data samples are obtained, the data samples are samples in the spatial domain, and their representation needs to be converted into the frequency domain.

[0042] S201: Based on the periodogram method of classical spectrum estimation, the spatial domain samples Perform spectrum estimation to obtain the time-frequency power spectrum of the road :

[0043]

[0044] S202: According to the road power spectrum density classification standard given by ISO8608, the state of each road sample is defined as the proportion of components from Class A to Class F. In the double logarithmic coordinate system, the displacement power spectrum of each level of road is a straight line. The smaller the spatial frequency, the greater the power spectrum density. In the same frequency range, the greater the power spectrum density, the greater the unevenness amplitude. When the spatial frequency is constant. The time frequency varies with the vehicle speed. Proportional to the spatial frequency power spectrum and the time-frequency power spectrum The relationship can be expressed as:

[0045]

[0046] S203: The sample space is limited. When defining the state, there will be multiple sample sequences with the same value but different positions. Without considering the phase, the amplitude of the spectrum estimate is equal, indicating that in a finite number of road point cloud data, the spectrum estimation results of sample sequences with the same value and different positions are the same. This type of sample is regarded as the same state, and the state space capacity formed by the sample is .

[0047] Furthermore, in the above S3, after converting the spatial domain samples into the frequency domain, all the frequency components are classified into road grades, and the proportions of A, B, C, D, E, and F roads are counted.

[0048] S301: Spectral estimation of the sample is performed to obtain ,right All frequency components in are classified.

[0049] S302: The density of one component is Then the distance of this component from each level is expressed as:

[0050]

[0051] Among them, the distance from the frequency to the boundary of each road grade is calculated using the Euclidean distance: , , , , , Use the minimum value The frequency component is attributed to the road class to which it belongs.

[0052] S303: The sample point cloud exists data, the frequency components of Class A roads in the sample can be summed up to , and then count each grade in the sample to get the proportion of each road grade in the sample. The components of each grade of road roughness after normalization It can be expressed as:

[0053] , to initialize the state.

[0054] Furthermore, in the above S4, after the basic state is obtained, under the condition that the Markov property is verified, the state space formed by the above S1, S2, and S3 is regarded as a Markov process.

[0055] S401: Form a state after state initialization , using the variable To indicate the time, The value range is , the initial time is set to 0, express The state of the Markov chain at that moment; All combinations of represent all possible states, which is a finite set, also known as the state space It is expressed as:

[0056]

[0057] S402: Since each state is independent of each other, the current state is , the next moment state is The extraction probability is:

[0058]

[0059] S403: The core property of Markov chain is that for any time , any current state And the next moment state is , and the possible state sequence at any previous time , all satisfy:

[0060]

[0061] S404: Summarizing S401, S402, and S403, extracting the state space formed by the road point cloud samples It can be regarded as a Markov process.

[0062] Furthermore, in the above S5, calculating the state transition matrix of any two states in the state space is the core task of the Markov chain model; this matrix describes the probability of transitioning from one state to another, and the transition probability converges to a stable value, indicating that the transition relationship between roads remains stable under the current state, that is, the relative roughness between roads of different levels maintains a certain relationship, which characterizes the overall roughness condition of the road.

[0063] S501: When initializing the transfer matrix, first create a The matrix of is the size of the state space, and each element of the matrix is ​​initialized to 0. This matrix is ​​called the transition frequency matrix, and each element represents the number of transitions from one state to another.

[0064] S502: Fill the matrix based on the collected data. When filling the state transition matrix, determining the transition frequency based on the similarity between states rather than simple counting can provide a more detailed and accurate model description, especially when dealing with continuous state spaces or when there are subtle differences between states. This method allows the model to capture the details of state changes, thereby providing richer information for analysis and prediction. This requires us to set a threshold. When the similarity between two states exceeds this threshold, the two states are considered "similar" and the state transition can be counted.

[0065] S503: Each time the slave state is observed To status The corresponding element in the matrix The value of is added by 1, and then the transition frequency matrix is ​​converted into a transition probability matrix. The specific operation is to divide each element by the sum of the elements in its row (that is, the sum of all transitions from this state). In this way, the sum of the elements in each row is 1, and the probability distribution is:

[0066] ,in, From the state To status The probability of From the state To status The number of observations, From the state The sum of all the outgoing transfers.

[0067] Furthermore, in the above S6, the validity and accuracy of the algorithm are verified using a labeled simulated point cloud dataset with 5% Gaussian white noise added. The following is a specific description of the verification embodiment:

[0068] S601: Use data labels to add 5% Gaussian white noise to the C-level road surface, and the sampling method of sample data is to randomly select road point cloud data.

[0069] S602: Perform 10 random samplings, extract 5000 points each time, obtain 10 samples, and perform discrete Fourier transform on the 10 data samples.

[0070] S603: Spectral estimation is performed on 10 data samples to initialize the state and form a Markov chain.

[0071] S604: Calculate the state transfer matrix according to the above Markov chain. The state transfer matrix is ​​used to describe the transition probability between each state in the Markov chain. Iterating the state transfer matrix means updating the state according to certain rules based on the current state and the transfer matrix so as to eventually converge to the target distribution, as shown in the attached figure. Figure 2 As shown in the figure, the probability of transition in the road ahead to C-class road is 52%, the probability of transition to B-class road is 15%, and the grade of the road ahead fluctuates between BC. The distribution of the state sequence of the Markov chain in a long time will be consistent with the target distribution, which shows the effectiveness of the method.

[0072] S605: To verify the accuracy of the model, 12,000 random road data are predicted from different sampling capacities. The accuracy test results are shown in the attached figure. Figure 2 As shown in the figure, out of a total of 262,144 data points, sampling 9,000 data points can keep the detection accuracy above 85%.

Claims

1. A MCMC road roughness detection method based on point cloud, characterized in that: Point cloud data is used for sampling to form samples, and the states of the samples are defined and initialized. Then, a Markov chain is constructed and the state transfer matrix is ​​calculated. Each state represents the proportion of different road levels in the sample. The calculation results of the state transfer matrix can be used to simulate and capture the transfer relationship between different levels of road roughness. The specific implementation steps are as follows: S1: Random selection of samples; S2: Definition of state; S3: initialization of state; S4: Construction of Markov chain; S5: calculation of state transfer matrix; S6: Validation of the method; Furthermore, in the above S1, the road point cloud is three-dimensional data, and each of its data points represents the roughness of the road in front of the vehicle; first, by sampling the point cloud, the roughness component analysis of the entire road is performed according to the sampling sequence, and the data sample is obtained by random sampling; Furthermore, in the above S2, after obtaining the data sample, the data sample is a sample in the spatial domain, and its expression is converted into the frequency domain according to the periodogram method of classical spectrum estimation; Furthermore, in the above S3, after converting the spatial domain samples into the frequency domain, all the frequency components are classified into road grades according to the road power spectrum density classification standard given by ISO8608, and the proportions of A, B, C, D, E, and F roads are counted; Furthermore, in the above S4, after the basic state is obtained, under the condition that the Markov property is verified, the state space formed by the above S1, S2, and S3 is regarded as a Markov process; Furthermore, in the above S5, calculating the state transition matrix of any two states in the state space is the task of the core part of the Markov chain model; this matrix describes the probability of transitioning from one state to another; the transition probability converges to a stable value, indicating that the transition relationship between roads remains stable in the current state, that is, the relative roughness between roads of different levels maintains a certain relationship, which helps to understand the overall condition of the road; Furthermore, in the above S6, a dataset of labeled simulated point cloud dataset with 5% Gaussian white noise added is used to verify the effectiveness and accuracy of the algorithm.

Images