Method for sequentially filling holes in pavement elevation map based on bilateral filtering

Abstract

The invention discloses a bilateral filtering-based pavement elevation map internal cavity sequence method, aims at improving the problem of low efficiency of an existing elevation map interpolation method, and can quickly and accurately complete map internal cavity repairing and filling. According to the method, an invalid grid neighborhood information quality function used for formulating a bilateral filtering filling sequence is constructed by combining the sparse degree and elevation variance of internal space points of neighborhood effective grids; parameters of an information quality function are optimized by adopting a positive gradient projection method; a self-adaptive repairing window is manufactured according to the macroscopic hole filling path; and sequentially filling the holes by using a bilateral filter with high real-time performance. And a brand-new mode is provided for filling the holes in the pavement elevation map.

Description

Technical Field

[0001] The invention relates to a method for sequentially filling holes in a road elevation map, in particular to a method for sequentially filling holes in a road elevation map based on bilateral filtering, and belongs to the field of 2.5D road reconstruction. Background Art

[0002] In recent years, the craze for driverless cars has triggered a series of technological innovations. Obtaining geographic data of roads is an important area in the field of autonomous driving. At present, in order to reduce costs and data storage scale, cameras are usually used as information acquisition sensors, and 2.5D elevation grid maps are used to store road surface information in front of the vehicle. However, cameras have problems such as low resolution at long distances and poor image quality in low-light environments. In addition, different types of cameras also have their own problems. The disparity map generated by the binocular camera through feature matching has white holes, and the scattering of outdoor light will cause the depth information obtained by the RGB-D camera to be missing. The above problems will eventually lead to holes in the elevation map, affecting the overall quality of the map, but the traditional elevation map interpolation method is inefficient and cannot meet the needs of daily vehicle driving. Therefore, it is necessary to propose a fast and accurate method for repairing holes in the elevation grid map to improve the map quality. Summary of the invention

[0003] The technical problem to be solved by the present invention is that the existing technology cannot quickly and accurately repair the internal holes of the elevation map, and thus a method for sequentially filling the internal holes of the road elevation map based on bilateral filtering is proposed.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A method for sequentially filling holes in road elevation maps based on bilateral filtering, the specific steps are as follows:

[0006] Step 1), combining the number of spatial points PN and the elevation variance σ 2 , construct the information quality function of a single valid grid;

[0007] Step 2), construct the weight functions ω1 and ω2 of the information quality function, and construct the weight difference function D in combination with the weight functions;

[0008] Step 3), randomly generate multiple groups of D parameters that meet the conditions, select a group of parameters that are closest to the optimization target as the initial parameters, and then use active gradient projection to further optimize the group of parameters;

[0009] Step 4), determining the effective area R of the invalid grid, and constructing its total information quality function based on the information quality function;

[0010] Step 5), according to the macro repair path, use reverse topology logic to make an adaptive repair sliding window;

[0011] Step 6), using a binocular camera to obtain road surface information, construct a corresponding three-dimensional point cloud and map it on an elevation grid map;

[0012] Step 7), using a bilateral filter to preprocess the initial elevation map of step 6) to eliminate discrete holes;

[0013] Step 8), converting the elevation map into a binary image according to whether it contains spatial points, extracting the internal hole contours of the binary image, and recording all contour positions;

[0014] Step 9), calculating the total information quality TIQ of each invalid grid, and sorting the priority of the invalid grids according to the value;

[0015] Step 10), use bilateral filter to sort TIQ The invalid grids are filled sequentially;

[0016] Step 11), step 9) to step 10) are repeated for q rounds;

[0017] Step 12), use a bilateral filter to fill the remaining invalid grids.

[0018] Preferably, in step 1), the information quality IQ of different effective grids satisfies the following conditions:

[0019] 1-1) When the number of internal spatial points is at a high order of magnitude, that is, when the number of spatial points PN>k2, the information quality IQ gap of different effective grids is determined by their elevation variance;

[0020] 1-2) When the number of internal spatial points is at a low order of magnitude, that is, when k1>PN>0, the information quality IQ gap of different effective grids is determined by the number of their spatial points;

[0021] 1-3) When the elevation variance difference of different effective grids is small, the information quality IQ of the effective grid with a high number of internal spatial points should be greater than the information quality IQ of the effective grid with a low number of internal spatial points.

[0022] Further preferably, in the step 1), k1 takes a value of 8 and k2 takes a value of 30.

[0023] Preferably, the information quality function in step 1) is:

[0024] IQ=ω1-ω2Tσ 2 (1)

[0025] In formula (1), ω1(lgPN) and ω2(lgPN) are weight functions of lgPN, and T is the amplification factor.

[0026] Preferably, in step 2), Tσ 2 The constant is 1, and the weight functions ω1, ω2 and the weight difference function D are constructed as functions of lgPN. The specific form is as follows:

[0027] ω1(lgPN)=a1 lg(PN)+b1 (2)

[0028] D=ω1(lgPN)-ω2(lgPN) (3)

[0029]

[0030] In formula (4), x is used to replace lgPN, D max is the maximum value of the weight difference function D, and a1, a2, b1, b2, c2, and e2 are the coefficients of the equation respectively.

[0031] Preferably, in step 3), the optimization object of the active gradient projection is the weight difference function D, the optimization target is to maximize (D(k2)-D(k1)), the boundary conditions need to ensure the continuity and smoothness of the curve, and the constraints include:

[0032] 3-1) The weight difference function D curve is monotonically increasing on [0, k2), so the parameters a1 and a2 are required to be greater than 0, and in order to avoid the curve changing too quickly, a1 is set to be no greater than 5;

[0033] 3-2) It is generally believed that the information quality provided by 8 valid neighborhood grids is greater than that of a single valid grid;

[0034] 3-3) When n is of low order, the information quality provided by n spatial point grids is lower than the sum of the information quality of n single spatial point grids. Therefore, the constraint is described by the following formula:

[0035]

[0036] In formula (5), n max It is the maximum number of spatial points that a single grid can contain. According to the elevation map resolution set to 1000, the formula (n*IQ(1,0)-IQ(n,0)) can take the minimum value when n=2.

[0037] Preferably, in step 4), the total information quality function is described as follows:

[0038]

[0039] In formula (6), TIQ(u,v) represents the total quality of the information in the area corresponding to the invalid grid (u,v), and IQ ij Represents the information quality of the effective grid (i, j); the effective neighborhood R is:

[0040] R={(i,j)|u-1≤i≤u+1,v-1≤j≤v+1,E(i,j)≠Nan,(i,j)≠(u,v)} (7)

[0041] In formula (7), E(i, j) is the elevation value of the valid grid (i, j), and Nan represents a null value.

[0042] Preferably, the topology grid coordinates determined by the reverse topology logic in step 5) are determined by the following formula:

[0043]

[0044] Preferably, in step 6), the binocular camera model is ZED2, the elevation map length is limited to 4m, and the resolution is 0.02m.

[0045] Preferably, in step 7) and step 12), the repair order of the bilateral filter is from top to bottom and from left to right.

[0046] Preferably, the algorithm used to extract the contour of the binary image in step 8) is the Canny algorithm.

[0047] Preferably, in step 11), the number of sequential filling rounds q is 10.

[0048] Compared with the prior art, the present invention has the following obvious outstanding technical effects and significant advantages:

[0049] 1) The present invention adopts a bilateral filter as a hole repair tool, which has higher real-time performance than traditional elevation map hole interpolation methods such as Kriging interpolation method and natural neighbor interpolation method. In addition, the method simultaneously considers the influence of pixel domain and spatial domain on the interpolation position, and can describe the relationship between invalid grid and neighborhood elevation, thus ensuring the accuracy of hole repair.

[0050] 2) Since the present invention determines the filling order of invalid grids by combining the number of spatial points contained in the effective grid of the elevation map and its elevation variance, the effective reference data used in each round of repair is of higher quality, which significantly improves the overall quality of the final repair result, and can undertake hole repair tasks in more scenarios (such as ramps, etc.), thereby enhancing the robustness and adaptability of the method;

[0051] 3) The present invention uses active gradient projection to optimize the parameters of the total information quality function of the invalid grid, so that the trend of the function curve can be more in line with expectations, and finally the filling order determined by the function is more reasonable, thereby improving the accuracy of hole repair;

[0052] 4) From a macroscopic point of view, the filling order of the present invention is from the edge of the hole to the center of the hole. According to this characteristic, an adaptive repair sliding window is made, which effectively increases the reference data that can be used by the bilateral filter and further improves the accuracy of hole repair;

[0053] 5) The present invention adopts the Canny algorithm for contour extraction. The Canny algorithm is a mature contour extraction algorithm with the advantages of fast speed and high precision. In addition, this method can determine the position of the hole contour and effectively reduce the sliding distance of the repair sliding window, thereby improving the real-time performance of the method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of a method for sequentially filling holes inside a road elevation map based on bilateral filtering according to the present invention.

[0055] Figure 2 It is the adaptive repair sliding window reverse topology logic diagram of the method of the present invention. DETAILED DESCRIPTION

[0056] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific examples of the present invention in conjunction with the accompanying drawings.

[0057] Embodiment 1

[0058] In this embodiment, a method for sequentially filling holes in a road elevation map based on bilateral filtering comprises the following steps:

[0059] Step 1), combining the number of spatial points PN and the elevation variance σ 2 , construct the information quality function of a single valid grid;

[0060] Step 2), construct the weight functions ω1 and ω2 of the information quality function, and construct the weight difference function D in combination with the weight functions;

[0061] Step 3), randomly generate multiple groups of D parameters that meet the conditions, select a group of parameters that are closest to the optimization target as the initial parameters, and then use active gradient projection to further optimize the group of parameters;

[0062] Step 4), determining the effective area R of the invalid grid, and constructing its total information quality function based on the information quality function;

[0063] Step 5), according to the macro repair path, use reverse topology logic to make an adaptive repair sliding window;

[0064] Step 6), using a binocular camera to obtain road surface information, construct a corresponding three-dimensional point cloud and map it on an elevation grid map;

[0065] Step 7), using a bilateral filter to preprocess the initial elevation map of step 6) to eliminate discrete holes;

[0066] Step 8), converting the elevation map into a binary image according to whether it contains spatial points, extracting the internal hole contours of the binary image, and recording all contour positions;

[0067] Step 9), calculating the total information quality TIQ of each invalid grid, and sorting the priority of the invalid grids according to the value;

[0068] Step 10), use bilateral filter to sort TIQ The invalid grids are filled sequentially;

[0069] Step 11), step 9) to step 10) are repeated for q rounds;

[0070] Step 12), use a bilateral filter to fill the remaining invalid grids.

[0071] This embodiment is based on the method of ordering holes inside road elevation maps using bilateral filtering, and aims to improve the low efficiency of existing elevation map interpolation methods, and can quickly and accurately complete the task of repairing and filling holes inside maps. This embodiment method combines the sparsity of internal spatial points of the neighborhood effective grids with the elevation variance to construct an invalid grid neighborhood information quality function for formulating the bilateral filtering filling order; an adaptive repair window is created based on the macroscopic hole filling path, providing a new way to fill holes inside road elevation maps.

[0072] Embodiment 2

[0073] This embodiment is basically the same as the first embodiment, except that:

[0074] In this embodiment, a method for sequentially filling holes in a road elevation map based on bilateral filtering comprises the following steps:

[0075] Step 1), combining the number of spatial points PN and the elevation variance σ 2 , construct the information quality function of a single valid grid; specifically:

[0076] The information quality IQ of different effective grids is set to meet the following conditions:

[0077] 1-1) When the number of internal spatial points is at a high order of magnitude, that is, when the number of spatial points PN>k2, the information quality IQ gap of different effective grids is determined by their elevation variance; where k2 is 30;

[0078] 1-2) When the number of internal spatial points is at a low order of magnitude, that is, when k1>PN>0, the information quality IQ gap of different effective grids is determined by the number of spatial points; where k1 takes the value of 8;

[0079] 1-3) When the elevation variance difference of different effective grids is small, the information quality IQ of the effective grid with a high number of internal spatial points should be greater than the information quality IQ of the effective grid with a low number of internal spatial points;

[0080] Step 2), construct the weight functions ω1 and ω2 of the information quality function, and construct the weight difference function D in combination with the weight functions;

[0081] Step 3), randomly generate multiple groups of D parameters that meet the conditions, select a group of parameters that are closest to the optimization target as the initial parameters, and then use active gradient projection to further optimize the group of parameters;

[0082] Step 4), determining the effective area R of the invalid grid, and constructing its total information quality function based on the information quality function;

[0083] Step 5), according to the macro repair path, use reverse topology logic to make an adaptive repair sliding window;

[0084] Step 6), using a binocular camera to obtain road surface information, construct a corresponding three-dimensional point cloud and map it on an elevation grid map;

[0085] Step 7), using a bilateral filter to preprocess the initial elevation map of step 6) to eliminate discrete holes;

[0086] Step 8), converting the elevation map into a binary image according to whether it contains spatial points, extracting the internal hole contours of the binary image, and recording all contour positions;

[0087] Step 9), calculating the total information quality TIQ of each invalid grid, and sorting the priority of the invalid grids according to the value;

[0088] Step 10), use bilateral filter to sort TIQ The invalid grids are filled sequentially;

[0089] Step 11), step 9) to step 10) are repeated for q rounds;

[0090] Step 12), use a bilateral filter to fill the remaining invalid grids.

[0091] The method for ordering internal holes in road elevation maps based on bilateral filtering in this embodiment combines the sparsity of internal spatial points in the neighborhood valid grids with the elevation variance to construct an invalid grid neighborhood information quality function for formulating the bilateral filtering filling order. The method in this embodiment aims to improve the low efficiency of existing elevation map interpolation methods and can quickly and accurately complete the task of repairing and filling holes in the map.

[0092] Embodiment 3

[0093] This embodiment is basically the same as the above embodiment, with the following features:

[0094] In this embodiment, if Figure 1 As shown, a method for sequentially filling holes inside a road elevation map based on bilateral filtering comprises the following steps:

[0095] Step S10: Combining the number of spatial points PN and the elevation variance σ 2 , construct the information quality function of a single effective grid; the information quality function can be constructed in the following form:

[0096] IQ=ω1-ω2Tσ 2 (1)

[0097] In formula (1), ω1(lgPN) and ω2(lgPN) are weight functions of lgPN, and T is the amplification factor.

[0098] Step S20: Construct the weight functions ω1 and ω2 of the information quality function, and construct the weight difference function D in combination with the weight functions. To facilitate the discussion of the relationship between the functions ω1 and ω2, first set Tσ 2 is constant to 1. The weight functions ω1, ω2 and the weight difference function D can be constructed as functions about lgPN, and the specific form is as follows:

[0099] ω1(lgPN)=a1lg(PN)+b1 (2)

[0100] D=ω1(lgPN)-ω2(lgPN) (3)

[0101]

[0102] In formula (4), x is used to replace lgPN, D max is the maximum value of the weight difference D. a1, a2, b1, b2, c2, e2 are the coefficients of the equation.

[0103] Step S30: randomly generate 10,000 groups of D parameters that meet the conditions, select a group of parameters that is closest to the optimization target as initial parameters, and then use active gradient projection to further optimize the group of parameters.

[0104] The optimization object of the active gradient projection is the weight difference function D, and the optimization goal is to maximize (D(k2)-D(k1)). The boundary conditions need to ensure the continuity and smoothness of the curve. The constraints include:

[0105] ① Curve D is monotonically increasing on [0, k2), so the parameters a1 and a2 are required to be greater than 0, and in order to avoid the curve changing too quickly, a1 is set to be no greater than 5;

[0106] ② It is generally believed that the information quality provided by 8 valid neighborhood grids is greater than that of a single valid grid;

[0107] ③ When n is of low order, the information quality provided by n spatial point grids is lower than the sum of the information quality of n single spatial point grids. In summary, the constraint condition can be described by the following formula:

[0108]

[0109] In formula (5), n max is the maximum number of spatial points that a single grid can contain. According to the elevation map resolution, it is set to 1000. The formula (n*IQ(1,0)-IQ(n,0)) can take the minimum value when n=2.

[0110] Step S40: Determine the valid region R of the invalid grid, and construct its total information quality function based on the information quality function. The total information quality function can be described by the following formula:

[0111]

[0112] In formula (6), TIQ(u,v) represents the total quality of the information in the area corresponding to the invalid grid (u,v), and IQ ij Represents the information quality of the effective grid (i, j). The effective neighborhood R is:

[0113] R={(i,j)|u-1≤i≤u+1,v-1≤j≤v+1,E(i,j)≠Nan,(i,j)≠(u,v)} (7)

[0114] In formula (7), E(i, j) is the elevation value of the valid grid (i, j), and Nan represents a null value.

[0115] Step S50: According to the macro-repair path, an adaptive repair sliding window is produced using reverse topology logic. Figure 2 The process and effects are demonstrated.

[0116] The topology grid coordinates determined by the reverse topology logic are determined by the following formula:

[0117]

[0118] Step S60: Use a binocular camera to obtain road surface information, construct a corresponding three-dimensional point cloud and map it on an elevation grid map. The binocular camera model in this embodiment is ZED2, the elevation map length is limited to 4m, and the resolution is 0.02m.

[0119] Step S70: Preprocess the initial elevation map of step S60 using a bilateral filter to eliminate discrete holes. During the preprocessing and finalization, the bilateral filter repairs in an order from top to bottom and from left to right.

[0120] Step S80: convert the elevation map into a binary image according to whether it contains a spatial point. Extract the internal hole contours of the binary image and record all contour positions. The algorithm used for contour extraction of the binary image is the Canny algorithm.

[0121] Step S90: Calculate the total information quality TIQ of each invalid grid, and sort the invalid grids by priority according to the value.

[0122] Step S100: Using bilateral filter to sort TIQ The invalid grids are filled sequentially.

[0123] Step S110: Steps S90 to S100 are repeated for q rounds. In this embodiment, the number of sequential filling rounds q is 10.

[0124] Step S120: Use a bilateral filter to fill the remaining invalid grids. The repair order of the bilateral filter is from top to bottom and from left to right.

[0125] The method of this embodiment uses a bilateral filter as a hole repair tool, which has higher real-time performance than traditional elevation map hole interpolation methods such as the Kriging interpolation method and the natural neighbor interpolation method. In addition, the method simultaneously considers the influence of the pixel domain and the spatial domain on the interpolation position, and can describe the relationship between the invalid grid and the neighborhood elevation, thereby ensuring the accuracy of hole repair. Since the filling order of the invalid grid is determined by combining the number of spatial points contained in the effective grid of the elevation map and the elevation variance thereof, the quality of the effective reference data used in each round of repair is higher, which significantly improves the overall quality of the final repair result, and can undertake hole repair tasks in more scenarios (such as ramps, etc.), thereby enhancing the robustness and adaptability of the method. The method of this embodiment uses an active gradient Projection optimizes the parameters of the total information quality function of the invalid grid so that the trend of the function curve can be more in line with expectations, and finally makes the filling order determined by the function more reasonable, thereby improving the accuracy of hole repair; the filling order of the method in this embodiment is from the edge of the hole to the center of the hole from a macro perspective. According to this feature, an adaptive repair sliding window is made, which effectively increases the reference data that can be used by the bilateral filter, and further improves the accuracy of hole repair; the method in this embodiment uses the Canny algorithm for contour extraction. The Canny algorithm is a mature contour extraction algorithm with the advantages of fast speed and high precision; and this method can determine the position of the hole contour, which can effectively reduce the sliding distance of the repair sliding window, thereby improving the real-time performance of the method of the present invention.

[0126] In summary, the method for sequencing internal holes in road elevation maps based on bilateral filtering in the present invention aims to improve the low efficiency of existing elevation map interpolation methods and can quickly and accurately complete the task of repairing and filling holes inside maps. The present invention constructs an invalid grid neighborhood information quality function for formulating the bilateral filtering filling sequence by combining the sparsity of internal spatial points of the neighborhood effective grid and the elevation variance; optimizes the parameters of the information quality function using the active gradient projection method; and creates an adaptive repair window based on the macroscopic hole filling path; and uses a bilateral filter with high real-time performance to sequentially fill holes. This provides a completely new way to fill holes inside road elevation maps.

[0127] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for sequentially filling holes in road elevation maps based on bilateral filtering, characterized in that: The specific steps are as follows: Step 1), combining the number of spatial points PN and the elevation variance σ 2 , construct the information quality function of a single valid grid; Step 2), construct the weight functions ω1 and ω2 of the information quality function, and construct the weight difference function D in combination with the weight functions; Step 3), randomly generate multiple groups of D parameters that meet the conditions, select a group of parameters that are closest to the optimization target as the initial parameters, and then use active gradient projection to further optimize the group of parameters; Step 4), determining the effective area R of the invalid grid, and constructing its total information quality function based on the information quality function; Step 5), according to the macro repair path, use reverse topology logic to make an adaptive repair sliding window; Step 6), using a binocular camera to obtain road surface information, construct a corresponding three-dimensional point cloud and map it on an elevation grid map; Step 7), using a bilateral filter to preprocess the initial elevation map of step 6) to eliminate discrete holes; Step 8), converting the elevation map into a binary image according to whether it contains spatial points, extracting the internal hole contours of the binary image, and recording all contour positions; Step 9), calculating the total information quality TIQ of each invalid grid, and sorting the priority of the invalid grids according to the value; Step 10), use bilateral filter to sort TIQ The invalid grids are filled sequentially; Step 11), step 9) to step 10) are repeated for q rounds; Step 12), use a bilateral filter to fill the remaining invalid grids.

2. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1 is characterized in that: In the step 1), the information quality IQ of different effective grids is set to meet the following conditions: 1-1) When the number of internal spatial points is at a high order of magnitude, that is, when the number of spatial points PN>k2, the information quality IQ gap of different effective grids is determined by their elevation variance; 1-2) When the number of internal spatial points is at a low order of magnitude, that is, when k1>PN>0, the information quality IQ gap of different effective grids is determined by the number of their spatial points; 1-3) When the elevation variance difference of different effective grids is small, the information quality IQ of the effective grid with a high number of internal spatial points should be greater than the information quality IQ of the effective grid with a low number of internal spatial points.

3. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 2 is characterized in that: In the step 1), k1 takes a value of 8 and k2 takes a value of 30.

4. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: The information quality function in step 1) is: IQ=ω1-ω2Tσ 2 (1) In formula (1), ω1(lgPN) and ω2(lgPN) are weight functions of lgPN, and T is the amplification factor.

5. The method for sequentially filling internal holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: In step 2), firstly Tσ 2 The constant is 1, and the weight functions ω1, ω2 and the weight difference function D are constructed as functions of lgPN. The specific form is as follows: ω1(lgPN)=a1 lg(PN)+b1 (2) D=ω1(lgPN)-ω2(lgPN) (3) In formula (4), x is used to replace lgPN, D max is the maximum value of the weight difference function D, and a1, a2, b1, b2, c2, and e2 are the coefficients of the equation respectively.

6. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: In the step 3), the optimization object of the active gradient projection is the weight difference function D, and the optimization target is to maximize (D(k2)-D(k1)). The boundary conditions need to ensure the continuity and smoothness of the curve. The constraints include: 3-1) The weight difference function D curve is monotonically increasing on [0, k2), so the parameters a1 and a2 are required to be greater than 0, and in order to avoid the curve changing too quickly, a1 is set to be no greater than 5; 3-2) It is generally believed that the information quality provided by 8 valid neighborhood grids is greater than that of a single valid grid; 3-3) When n is of low order, the information quality provided by n spatial point grids is lower than the sum of the information quality of n single spatial point grids. Therefore, the constraint is described by the following formula: In formula (5), n max It is the maximum number of spatial points that a single grid can contain. According to the elevation map resolution set to 1000, the formula (n*IQ(1,0)-IQ(n,0)) can take the minimum value when n=2.

7. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: In step 4), the total information quality function is described as follows: In formula (6), TIQ(u,v) represents the total quality of the information corresponding to the invalid grid (u,v), IQ ij Represents the information quality of the effective grid (i, j); the effective neighborhood R is: R={(i,j)|u-1≤i≤u+1,v-1≤j≤v+1,E(i,j)≠Nan,(i,j)≠(u,v)} (7) In formula (7), E(i, j) is the elevation value of the valid grid (i, j), and Nan represents a null value.

8. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: The topology grid coordinates determined by the reverse topology logic in step 5) are determined by the following formula:

9. The method for sequentially filling holes in a road elevation map based on bilateral filtering according to claim 1, characterized in that: In the step 6), the binocular camera model is ZED2, the elevation map length is limited to 4m, and the resolution is 0.02m.

10. The method for sequentially filling holes in road elevation maps based on bilateral filtering according to claim 1, characterized in that: In step 7) and step 12), the repair order of the bilateral filter is from top to bottom and from left to right.

Images