A method for lidar point cloud radius filtering based on genetic algorithm

Abstract

A lidar point cloud radius filtering method based on genetic algorithm relates to the technical field of lidar data processing. The present invention is to solve the problems that the parameter determination method of traditional filtering methods is time-consuming, and the noise recall rate is low in complex scenarios, and the details in the point cloud image will be lost, resulting in low processing efficiency. The point cloud noise reduction method based on genetic algorithm is used to calculate the optimal radius and adjacent point threshold according to the characteristics of the point cloud data, reduce the steps of manual verification of parameters, and can improve the noise recall rate while ensuring the noise reduction accuracy rate and the origin retention rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of lidar data processing. Background Art

[0002] With the wide application of lidar technology in the fields of autonomous driving, environmental perception, etc., the generated lidar point cloud data is often affected by various interferences, such as noise, stray points, etc. These interferences affect the quality and reliability of lidar data. Therefore, filtering methods are needed to improve the data processing efficiency. Existing traditional filtering methods include statistical filtering, radius filtering, Gaussian filtering, etc. These traditional methods have continuously improved the accuracy and efficiency of point cloud denoising, providing strong support for point cloud applications in various fields. Scholars have also conducted some research based on these traditional methods. For example: Chen et al. first planarize the point cloud, separate it from the main body point cloud, and then use the remaining point cloud for mean statistical filtering, which can eliminate the influence of part of the plane on the denoising effect and remove the discrete point cloud around the main body, but the internal noise is not well removed. Yuan Hua et al. proposed a bilateral filtering point cloud denoising algorithm based on noise classification for the problem of different scale noises in the three-dimensional point cloud data model. Zhou et al. divided point cloud denoising into two stages of large and small thresholds to solve the problem of time-consuming traditional denoising methods. Each stage uses hierarchical processing, effectively improving the denoising speed, but the error removal of the target point cloud is relatively large. Liu Bin et al. proposed an adaptive double-radius filtering algorithm applicable to complex scenes and various scale noises, overcoming the defect that radius filtering cannot identify small-scale noises and realizing adaptive filtering of multi-scale noises.

[0003] However, the amount of point cloud data is generally relatively large, and the parameter input of traditional filtering methods needs to be continuously tried manually to obtain the best parameters, resulting in a waste of a lot of time to determine the parameter selection; and in complex scenes, the noise recall rate after traditional filtering is not very high, and some details in the point cloud image will be lost, resulting in low processing efficiency. Summary of the Invention

[0004] The present invention is to solve the problems that the parameter determination method of traditional filtering methods is time-consuming, and the noise recall rate is low in complex scenes, and details in the point cloud image will be lost, resulting in low processing efficiency. Now, a lidar point cloud radius filtering method based on a genetic algorithm is provided.

[0005] A lidar point cloud radius filtering method based on a genetic algorithm includes the following steps:

[0006] Step 1: Use each individual value in the population R k to perform radius filtering on the point cloud to be filtered respectively, and obtain the filtered point cloud corresponding to different individuals. The value of the individual is the filtering radius, and k represents the number of iterations;

[0007] Step 2: Calculate the fitness of each individual according to the filtered point cloud corresponding to each individual in population R k Select the individual with the maximum fitness as the optimal individual and put the optimal individual into the optimal individual set;

[0008] Step 3: Determine whether k is less than the maximum number of iterations Z. If yes, execute Step 5; otherwise, execute Step 4;

[0009] Step 4: Use the value of the individual with the maximum fitness in the optimal individual set to perform radius filtering on the point cloud to be filtered, and complete the radius filtering of the lidar point cloud;

[0010] Step 5: Normalize the fitness of each individual in population R k to probability values, and select excellent individuals by roulette according to the probability values of each individual to form an excellent population

[0011] Step 6: Encode the values of the excellent individuals in the excellent population into binary strings, perform crossover and mutation operations on the binary strings of the excellent individuals in sequence to obtain new binary strings, decode the new binary strings to obtain the values of new individuals, and add the new individuals to the excellent population to obtain the updated population R k+1 k+1 , set k = k + 1, and return to Step 1.

[0012] Further, the above-mentioned radius filtering of the point cloud to be filtered includes:

[0013] Calculate the number of neighborhood points of each data point in the point cloud to be filtered within the filtering radius r i k range, where r i k is the value of the i-th individual in population R k ;

[0014] Calculate the neighborhood point number threshold N θ ;

[0015] When the number of neighborhood points of a data point is greater than or equal to the neighborhood point number threshold N θ , then the data point is a core point,

[0016] When the number of neighborhood points of a data point is less than the neighborhood point number threshold N θ and greater than 0, then the data point is a boundary point,

[0017] When the number of neighborhood points of a data point is equal to 0, then the data point is a noise point,

[0018] Delete the noise points in the point cloud to be filtered to achieve radius filtering of the point cloud.

[0019] Further, calculating the neighborhood point number threshold N according to the number of neighborhood points described above θ , includes:

[0020]

[0021] Among them, β is a threshold parameter and its value is 0.3, N is the total number of all data points in the point cloud to be filtered, N n (r i k ) is the number of neighborhood points of the nth point in the point cloud to be filtered within the radius r i k range.

[0022] Further, calculating the fitness of each individual includes:

[0023] Calculate the signal-to-noise ratio of the filtered point cloud corresponding to the ith individual in the population R k according to the following formula

[0024]

[0025] Among them, and are respectively the number of signal points and noise points in the filtered point cloud corresponding to the ith individual in the population R k ;

[0026] Calculate the fitness F of the ith individual in the population R k according to the following formula i k :

[0027]

[0028] Further, normalizing the fitness of each individual in the population R k to a probability value includes:

[0029] Normalize the fitness F of the ith individual in the population R k to a probability value according to the following formula i k :

[0030]

[0031] Among them, I is the total number of individuals in the population R k , is the probability value of the ith individual in the population R k .

[0032] Further, after the binary strings corresponding to the excellent individuals are successively subjected to crossover and mutation processing, the new binary strings are obtained, including:

[0033] Randomly select multiple individuals from the excellent population according to the crossover rate, and pair up the selected individuals two by two to form a crossover pair;

[0034] Randomly set the crossover positions of each crossover pair, and exchange the crossover positions of the binary strings of the two individuals in the same crossover pair to obtain the binary strings of the individuals after crossover;

[0035] Randomly set the mutation positions of each individual, and perform the operation of taking the opposite gene value on the mutation positions of the binary strings of the individuals after crossover to obtain the binary strings of the individuals after mutation as the new binary strings.

[0036] Further, the above iteration times k = 0, 1, 2,..., Z,

[0037] When k = 0, the values of the individuals in the initial population R 0 are randomly generated,

[0038] The initial population R 0 The initial value r i 0 of the i-th individual in it has the value range expressed as:

[0039] a ≤ r i 0 ≤ b × max{D euclidean},

[0040] where a is the lower bound of the value range of r i 0 , b is the upper bound coefficient of the value range of r i 0 , D euclidean is the Euclidean distance between the point Q m and the point Q n :

[0041]

[0042] (X m , Y m , Z m ) and (X n , Y n , Z n ) are the three-dimensional coordinates of the point Q m and the point Q n respectively.

[0043] Further, screening out excellent individuals by roulette according to the probability values of each individual includes:

[0044] Determining the number of excellent individuals screened out in population R according to a preset selection probability k and screening out excellent probability values from the probability values of all individuals in population R by roulette based on the number of excellent individuals. The individuals corresponding to the excellent probability values are excellent individuals.

[0045] k

[0046]

[0046] Further, encoding the values of the excellent individuals in the excellent population into binary strings includes:

[0047] Mapping the value of each excellent individual to the integer range, then multiplying by an order of magnitude unit, and finally converting the obtained integer into a binary string to complete the encoding.

[0048] Further, decoding the new binary string to obtain the value of the new individual includes:

[0049] Converting the new binary string into an integer, and then dividing the integer by the order of magnitude unit to obtain the value of the new individual.

[0050] A lidar point cloud radius filtering method based on a genetic algorithm according to the present invention uses a point cloud noise reduction method based on a genetic algorithm, calculates the optimal radius and the adjacent point threshold according to the characteristics of the point cloud data, reduces the steps of manual verification of parameters, and can improve the noise recall rate while ensuring the noise reduction accuracy rate and the origin retention rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flowchart of a lidar point cloud radius filtering method based on a genetic algorithm described in the specific implementation manner;

[0052] Figure 2 is a selection and crossover flowchart;

[0053] Figure 3 is a fitness value curve graph of the rabbit model in each generation;

[0054] Figure 4 is an optimal individual curve graph of the rabbit model in each generation;

[0055] Figure 5 is a point cloud graph of the noisy rabbit model;

[0056] Figure 6 is a point cloud graph of the point cloud of the noisy rabbit model after traditional radius filtering;

[0057] Figure 7 ​It is the point cloud map after statistical filtering of the noisy bunny model point cloud;

[0058] Figure 8 It is the point cloud map after radius filtering of the noisy bunny model point cloud as described in the specific implementation;

[0059] Figure 9 It is the point cloud map of the noisy residential area;

[0060] Figure 10 It is the point cloud map after traditional radius filtering of the noisy residential area point cloud;

[0061] Figure 11 It is the point cloud map after statistical filtering of the noisy residential area point cloud

[0062] Figure 12 It is the point cloud map after radius filtering of the noisy residential area point cloud as described in the specific implementation. Specific implementation

[0063] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0064] Specific implementation one: Refer to Figure 1 This specific implementation will be specifically described. A lidar point cloud radius filtering method based on a genetic algorithm described in this specific implementation is as follows:

[0065] 1. Construct a fitness function

[0066] A high signal-to-noise ratio is an indicator of good signal quality. If the signal-to-noise ratio is positive, it means that the signal power is greater than the noise power; otherwise, it means that the signal power is less than the noise power. Therefore, first construct the signal-to-noise ratio R of the point cloud SN :

[0067]

[0068] Among them, P signal and P noise are the numbers of signal points and noise points in the point cloud respectively.

[0069] In order to obtain good filtering results, in addition to a high signal-to-noise ratio requirement, high precision is also required, that is, as few truly informative signals as possible should be filtered out. Therefore, calculate the fitness F of the individuals in the population fitness as follows:

[0070] F fitness = R SN × P signal 。

[0071] 2. Construct the initial population:

[0072] Generate an initial value for each individual in the population of the genetic algorithm in each dimension to represent the filtering radius of the point cloud. Among them, the initial value (initial filtering radius) r 0 of the i-th individual in the initial population R i 0 has a value range of:

[0073] a ≤ r i 0 ≤ b × max{D euclidean},

[0074] where a is the lower bound of the value range of r i 0 , b is the upper bound coefficient of the value range of r i 0 , and a and b are determined according to the order of magnitude unit of the three-dimensional coordinates.

[0075] D euclidean is the Euclidean distance between the point Q m (X m , Y m , Z m ) and the point Q n (X n , Y n , Z n ):

[0076]

[0077] Take a random number between 0 and 1, and perform a linear mapping of this random number with the above value range to ensure that each dimension of each individual in the population is randomly distributed within the specified range.

[0078] 3. Genetic algorithm iteration:

[0079] Let the current iteration number be k, k = 1, 2,..., Z, where Z is the maximum number of iterations.

[0080] S1: Use each individual in the population R k at the current iteration number k to perform radius filtering on the point cloud to be filtered, and obtain the filtered point clouds corresponding to different individuals.

[0081] Specifically, the above radius filtering process is as follows: For each data point in the point cloud, calculate the number of neighboring points within the filtering radius, and calculate the neighborhood point number threshold N based on the number of neighboring points θ , if the number of neighboring points is greater than or equal to the predetermined neighborhood point number threshold N θ , then mark the data point as a core point. If the number of neighboring points is less than the threshold but greater than 0, the data point is marked as a boundary point. If the number of neighboring points is 0, the data point is marked as a noise point. Remove the data points marked as noise points from the point cloud, and retain the core points and boundary points to complete the radius filtering. The advantage of this process is automatic optimization, which can obtain optimized parameters suitable for different scenarios and improve the point cloud denoising effect

[0082] The above neighborhood point number threshold N θ is obtained through the following formula

[0083]

[0084] where β is the threshold parameter with a value of 0.3. N is the total number of all data points in the point cloud, and N n (r i k ) is the number of neighboring points of the nth point in the point cloud to be filtered within the radius r i k .

[0085] S2: Calculate the signal-to-noise ratio of each point cloud at the current iteration number k according to the following formula

[0086]

[0087] where and are respectively the number of signal points and noise points in the point cloud corresponding to the i-th individual in the population R k at the current iteration number k

[0088] S3: Calculate the fitness F k of the i-th individual in the population R i k at the current iteration number k according to the following formula

[0089]

[0090] S4: Take the individual with the maximum fitness in the population R k as the optimal individual and put it into the optimal individual set

[0091] S5: Determine whether k is less than the maximum iteration number Z. If yes, execute S7; otherwise, execute S6

[0092] S6: Take the individual with the largest fitness value in the optimal individual set as the final individual, and use the value of this final individual to perform radius filtering on the point cloud to complete the radius filtering of the lidar point cloud.

[0093] S7: Normalize the fitness of each individual in population R k to a probability value

[0094]

[0095] where I is the total number of individuals in population R at the current iteration number k k and is the probability value of the i-th individual in population R at the current iteration number k k .

[0096] S8: Determine the number of excellent individuals to be selected from population R according to the preset selection probability. Use the roulette wheel method to select excellent probability values from the probability values of all individuals in population R k . The individuals corresponding to the excellent probability values are excellent individuals. Use all excellent individuals to form the excellent population at the current iteration number k k . The number of excellent individuals is determined by the preset selection rate.

[0097] S9: Encode each excellent individual in the excellent population . That is: Map the value (filtering radius) of each excellent individual at the current iteration number k to the integer range, then multiply by an order of magnitude unit, and then convert the obtained integer into a binary code.

[0098] S10: Perform crossover and mutation on the encoded individuals in the excellent population in sequence to obtain new individuals.

[0099] The way of the crossover is as follows:

[0100] In the excellent population , randomly select multiple crossover pairs according to the crossover rate, where each crossover pair includes two excellent individuals. Randomly determine the crossover positions of each crossover pair, and exchange the values at the crossover positions of the two excellent individuals in the same crossover pair to obtain two crossover individuals. In this way, the exchange of gene information of individuals is realized, which promotes the gradual optimization of the population during the evolution process and ensures the transmission of gene information and the maintenance of diversity.

[0101] The way of the mutation is as follows:

[0102] Randomly determine the mutation positions of each individual, and perform the operation of taking the inverse gene value at the mutation positions of each crossover individual to obtain the mutated individuals.

[0103] Finally, all mutant individuals are added to the elite population to obtain the updated population R k+1 .

[0104] The randomness of the above selection, crossover, and mutation makes the individual gene combinations more diverse in the population, increasing the exploration ability of the population. This diversity helps the genetic algorithm avoid falling into local optimal solutions and is more conducive to finding the global optimal solution.

[0105] S11: Decode the new individual, that is: convert the new binary string into an integer, and then divide the integer by the order of magnitude unit to obtain the value of the new individual. Then add the decoded new individual to the elite population to form the updated population R k+1 . Finally, make k = k + 1 and return to S1.

[0106] Experimental verification is carried out on a lidar point cloud radius filtering method based on the genetic algorithm described in the above specific implementation manners as follows:

[0107] First, a rabbit model in the Stanford database is used for simulation analysis. Gaussian noise points with a mean of 0 and a standard deviation of 0.1 are added to the original point cloud to form a noisy point cloud. Then, the publicly available real residential area point cloud provided by NSF (National Science Foundation, United States) is used for testing. In the experiment, the point cloud data only includes position information.

[0108] The optimal four groups of parameter results of the rabbit model in radius filtering and statistical filtering obtained through multiple tests are shown in Table 1 and Table 2 respectively as follows:

[0109] Table 1 Radius filtering parameters

[0110]

[0111]

[0112] Table 2 Statistical filtering parameters

[0113]

[0114] In radius filtering, as the radius increases, the number of filtered noise points gradually decreases, and at the same time, the signal-to-noise ratio is lower; while in statistical filtering, as the distance threshold increases, the number of filtered noise points also decreases, and at the same time, the signal-to-noise ratio is lower. The parameters with similar signal-to-noise ratios are selected for radius filtering and statistical filtering, that is, the parameters with SNR of 30.2846 and 30.4384 respectively. The results of the Stanford rabbit model in each filtering algorithm are shown in Table 3:

[0115] Table 3 Results of the rabbit model with various filtering algorithms

[0116]

[0117] The noisy point cloud map and the maps after filtering by various algorithms are shown as Figure 5 、 Figure 6 、 Figure 7 、 Figure 8 . According to what is shown in the figure, there is still a little noise after the statistical filtering is completed, while both the radius filtering and the proposed algorithm have good noise reduction effects.

[0118] To further verify the effect of the algorithm in complex scenarios, a section of real lidar point cloud data of a residential area provided by NSF is selected. The noisy point cloud is shown as follows. The number of point clouds is 31,870. The results after noise reduction by various algorithms are shown in Table 4 and Appendix Figure 9 、 Figure 10 、 Figure 11 、 Figure 12 .

[0119] Table 4 Results of various filtering algorithms in the residential area

[0120]

[0121]

[0122] N q is the number of removed noises; N g is the total number of noises; N y is the number of removed points; N o is the number of points in the remaining filtered point cloud that do not belong to noises; N f is the number of remaining filtered points.

[0123] As can be seen from the attached drawings, in complex scenarios, compared with the statistical filtering, the radius filtering has less drift noise, but more signal points are also filtered out, and the contours of the trees are cut. While the proposed algorithm can better preserve the contours of the houses and trees while filtering the noise, and reduce the number of drift noises.

[0124] The denoising precision P d , the noise recall rate R d , and the origin retention rate R o are introduced to quantitatively compare the denoising degree of the filtering algorithms. The evaluation indexes of noise reduction by various algorithms for the Stanford rabbit model are shown in Table 5.

[0125] Table 5 Evaluation indexes of noise reduction results for the rabbit model

[0126]

[0127] The noise reduction precision rate of radius filtering and statistical filtering is high, indicating that there are fewer noise points in the filtered point cloud. However, the recall rate is relatively low, indicating that some noise points are not correctly identified and filtered. The algorithm accuracy rate and the original point retention rate of this embodiment are high, and the recall rate has also been improved. Therefore, in this embodiment, radius filtering can be achieved and better noise reduction can be performed.

[0128] The evaluation index results of noise reduction for each algorithm in the residential area are shown in Table 6.

[0129] Table 6 Evaluation Index of Noise Reduction Results in Residential Area

[0130]

[0131] In the residential scene, due to the more complex target objects and noise types, when the number of noises is large enough, it is easy to mix with the main body, resulting in a significant decrease in the noise recall rate compared to the rabbit model. However, this embodiment is about 16% higher than radius filtering and statistical filtering. The noise reduction precision rate is slightly lower than that of radius filtering and is not much different from that of statistical filtering. The original point retention rates of the three algorithms are also not much different.

[0132] Based on the above evaluation indexes reflecting the noise reduction effect, this embodiment can complete the noise reduction task without repeatedly testing parameters and can improve the noise recall rate to a certain extent. While ensuring the noise reduction precision rate and the original point retention rate, the noise recall rate in a simple scene is slightly increased; in a complex scene, it is about 21% higher than traditional radius filtering and about 16% higher than statistical filtering. This embodiment has important application value for improving the quality of lidar imaging.

[0133] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A lidar point cloud radius filtering method based on a genetic algorithm, characterized in that Including the following steps: Step 1: Use the population R k to perform radius filtering on the point cloud to be filtered respectively with the value of each individual in it, and obtain the filtered point clouds corresponding to different individuals. The value of the individual is the filtering radius, and k represents the number of iterations; Step 2: According to the population R k Calculate the fitness of each individual based on the filtered point cloud corresponding to each individual in it, take the individual with the maximum fitness as the optimal individual, and put the optimal individual into the optimal individual set; Calculating the fitness of each individual includes: Calculate the population R according to the following formula k The signal-to-noise ratio of the filtered point cloud corresponding to the i-th individual in Among them, and are the numbers of signal points and noise points in the filtered point cloud corresponding to the \(i\)-th individual in population \(R\), k respectively. Calculate the population R according to the following formula k The fitness F of the i-th individual in i k :[[]]END]] Step 3: Determine whether k is less than the maximum number of iterations Z. If so, execute Step 5; otherwise, execute Step 4; Step 4: Use the value of the individual with the maximum fitness in the optimal individual set to perform radius filtering on the point cloud to be filtered, and complete the radius filtering of the lidar point cloud; Step Five: Normalize the fitness of each individual in population R k into probability values, and select excellent individuals by roulette according to the probability values of each individual to form an excellent population Step 6: Encode the values of the excellent individuals in the excellent population into binary strings. After performing crossover and mutation operations on the binary strings of the excellent individuals in sequence, obtain new binary strings. Decode the new binary strings to obtain the values of the new individuals, and add the new individuals to the excellent population to obtain the updated population R k+1 , let k = k + 1, and return to Step 1.

2. The method for filtering the radius of lidar point cloud based on genetic algorithm according to claim 1, characterized in that Performing radius filtering on the point cloud to be filtered includes: Calculate the number of neighborhood points of each data point in the point cloud to be filtered within the filtering radius r i k range, r i k is the value of the i-th individual in the population R k ; Calculate the neighborhood point quantity threshold N according to the quantity of the neighborhood points θ ; When the number of neighborhood points of a data point is greater than or equal to the neighborhood point number threshold N θ then this data point is a core point When the number of neighborhood points of a data point is less than the neighborhood point number threshold N θ and greater than 0, then the data point is a boundary point When the number of neighborhood points of a data point is equal to 0, then the data point is a noise point. Delete the noise points in the point cloud to be filtered to achieve the radius filtering of the point cloud.

3. The method for filtering the radius of lidar point cloud based on genetic algorithm according to claim 2, wherein Calculating a neighborhood point quantity threshold N according to the quantity of the neighborhood points θ , comprising: Among them, β is a threshold parameter with a value of 0.3, N is the total number of all data points in the point cloud to be filtered, N n (r i k ) is the number of neighboring points of the nth point in the point cloud to be filtered within the radius r i k range.

4. A lidar point cloud radius filtering method based on genetic algorithm according to claim 1, characterized in that Said population R k Normalizing the fitness of each individual in it to a probability value, including: Normalize the fitness F of the i-th individual in population R according to the following formula: k where the fitness F i k is normalized to a probability value: Among them, I is the total number of individuals in population R k where is the probability value of the i-th individual in population R k .

5. A lidar point cloud radius filtering method based on a genetic algorithm according to claim 1 or 4, characterized in that After sequentially performing crossover and mutation processing on the binary strings corresponding to the excellent individuals, obtaining new binary strings includes: According to the crossover rate, multiple individuals are randomly selected from the elite population and the selected individuals are paired up into a crossover pair; Randomly set the crossover positions of each crossover pair, and exchange the crossover positions of the binary strings of the two individuals in the same crossover pair to obtain the binary string of the individual after crossover; Randomly set the mutation positions of each individual, and perform the operation of taking the inverse gene value on the mutation positions of the binary strings of the individuals after crossover to obtain the binary string of the individual after mutation as the new binary string.

6. A lidar point cloud radius filtering method based on a genetic algorithm according to claim 1, 2 or 4, characterized in that The number of iterations k = 0, 1, 2,..., Z. When k = 0, the values of individuals in the initial population R 0 are randomly generated, Initial population R 0 The initial value r of the i-th individual in i 0 is expressed as the value range of: a ≤ r i 0 ≤ b × max{D euclidean}, where a is the lower bound of the value range of r i 0 and b is the upper bound coefficient of the value range of r i 0 , D euclidean is the Euclidean distance between the point Q m in the point cloud to be filtered and the point Q n : (X m , Y m , Z m ) and (X n , Y n , Z n ) are the three-dimensional coordinates of point Q m and point Q n , respectively.

7. A lidar point cloud radius filtering method based on a genetic algorithm according to claim 1, characterized in that Selecting excellent individuals in the form of roulette according to the probability values of each individual includes: Determine the number of excellent individuals selected from the population R according to the preset selection probability k in the population R Select excellent probability values from the probability values of all individuals in population R by roulette wheel method based on the number of excellent individuals. The individuals corresponding to the excellent probability values are excellent individuals. k All individuals' probability values are screened to obtain excellent probability values, and the individuals corresponding to the excellent probability values are excellent individuals.

8. A lidar point cloud radius filtering method based on a genetic algorithm according to claim 1, characterized in that Said excellent population Encoding the values of excellent individuals in the excellent population into binary strings, including: Map the value of each excellent individual to the integer range, then multiply it by an order of magnitude unit, and finally convert the obtained integer into a binary string to complete the encoding.

9. A method for filtering the radius of lidar point cloud based on genetic algorithm according to claim 8, characterized in that, Decoding the new binary string to obtain the value of the new individual includes: Convert the new binary string into an integer, and then divide the integer by the order of magnitude unit to obtain the value of the new individual.

Images