CN110285805A - An Adaptive Interpolation/Segmentation Processing Method for Data Holes - Google Patents

Abstract

The invention provides a data hole self-adaptive interpolation/segmentation scheme, which can be used for three-dimensional elevation matching of REM containing data holes, and belongs to the field of terrain aided navigation. Since data holes are unavoidable in actual matching, the present invention proposes an effective solution. The invention first includes classifying the DEM by using the error rate of the reference map, extracting the ROI in the DEM; then using the interpolation error rate to calculate the interpolation error between the REM and the ROI, and classifying the size of the data hole according to the interpolation error; finally using the self-adaptive interpolation The /segmentation strategy processes the REM to reduce the impact of data holes in the actual matching on the 3D elevation matching.

Description

An Adaptive Interpolation/Segmentation Processing Method for Data Holes

technical field

The invention provides a data hole adaptive interpolation/segmentation processing method, which can be used for three-dimensional elevation matching of a real-time elevation map (Real-time Elevation Map, REM) containing data holes, and belongs to the field of terrain-assisted navigation.

Background technique

The fully autonomous terrain-assisted navigation system can be widely used in complex electromagnetic environments, deep space exploration and underwater fields, and plays a unique role in the national economy and national defense construction that cannot be replaced by satellite navigation or other navigation. Compared with the traditional single-point and one-dimensional sequence matching, in the 3D terrain matching based on 3D elevation sampling, the sampling points are greatly increased, which greatly reduces the probability of false matching caused by terrain similarity; Unlike the sampled real-time map, where the pixel values are all gray values, the 3D terrain matching is all terrain elevation, and the matching performance is better in areas with significant terrain fluctuations. Therefore, 3D elevation matching has unique advantages in high-precision autonomous navigation. and broad application prospects.

The basic working principle of 3D terrain-assisted navigation is shown in Figure 1. First, the terrain in the matching area is digitized, and a reference map database based on Digital Elevation Model (DEM) is constructed and stored in the navigation computer; When matching the terrain area, use the three-dimensional elevation measurement sensor to measure the terrain REM at the position of the carrier; then, as shown in Figure 2, take the current position (i, j) of the inertial navigation system (Inertial Navigation System, INS) as the center, According to the larger value σ of the estimated position error of the INS in the longitude and latitude directions, according to the 3σ criterion, determine the area to be matched in the reference map database. The area to be matched contains a total of I×J DEMs, and the center coordinates are (i, j) The DEM of DEM is recorded as DEM _(i,j) ; finally, in the area to be matched, the REM is matched with the DEM reference map contained in the area to be matched, and the matching position is obtained, and the matching position is fed back to the INS to correct Cumulative error of INS.

Commonly used 3D elevation measurement sensors include Interferometric Synthetic Aperture Radar (InSAR), LiDAR (Light Detection and Ranging, LiDAR), stereo vision cameras, ultrasonic rangefinders, and infrared rangefinders. However, these three-dimensional elevation measurement sensors are prone to data holes when acquiring REM in real time, which seriously affects the matching performance and usability of terrain-assisted navigation systems. The present invention takes InSAR as an example to introduce the causes of data holes.

InSAR measurement is a recently developed three-dimensional terrain elevation measurement technology, which is the application extension and expansion of Synthetic Aperture Radar (SAR). InSAR measurement technology uses two SAR images of the same area as the basic processing data, and obtains a three-dimensional terrain elevation image through interference and phase unwrapping. InSAR measurement can work around the clock and all day, with a large surveying range and high data processing efficiency.

However, as shown in Figure 3, in SAR images, due to the geometric relationship between imaging and ground scenes, overlapping and shadowing are common phenomena, especially in areas with large terrain elevation fluctuations such as mountainous areas or urban areas. Overlapping and shadow areas correspond to areas that cannot be unwrapped on the phase map, resulting in data loss in the REM and data holes, as shown in Figure 4, where the total number of normal points in the REM data is P, and the number of data holes is M, the number of missing data for the mth data hole is N _m .

In topographic mapping, multi-angle measurement is used, and the complementarity between multiple maps is used to eliminate the influence of data holes. However, since the carrier using terrain-assisted navigation only passes through the target area once, multi-angle measurement cannot be performed, and the navigation performance of terrain-assisted navigation is better in areas with large terrain elevation fluctuations. Therefore, in terrain-assisted navigation based on InSAR measurement It is easy to generate data holes. In the existing related research, the focus is on the feature extraction method, fast calculation algorithm and matching algorithm of the 3D elevation map, but the problem of data holes in the 3D elevation measurement has not been studied. Therefore, it is of great application value to study a data hole adaptive interpolation/segmentation processing method to reduce the impact of data holes on the performance of 3D elevation terrain matching.

Contents of the invention

The technology of the present invention solves the problem: adopts the interpolation/segmentation strategy, uses the error rate of the reference map and the interpolation error rate to classify data holes, constructs an adaptive interpolation/segmentation method for data holes, and improves the performance of terrain-assisted navigation.

Technical points of the present invention:

The accuracy of REM is mainly determined by the acquisition error of the original data and the elevation interpolation error. Among them, the data acquisition error mainly includes the error of the three-dimensional elevation measurement sensor, the installation error and the data processing error, etc. The elevation interpolation error is the difference between the interpolation point and the actual measured elevation. deviation between. The invention adopts the error rate of the reference image and the error rate of the interpolation to represent the sizes of the two types of errors respectively, and provides a basis for the self-adaptive interpolation/segmentation method of the data hole.

1. Reference image error rate

Baseline error rate It is defined as follows:

in: Indicates the base map error rate between REM and DEM ^{(i, j)} reference map, the total number of normal data in REM is P; f _REM (p) is the REM elevation value of the pth data normal place; f _DEM (p) is the DEM elevation value at the normal position of the pth data corresponding to the REM.

Traverse the area to be matched, the traversal method is shown in Figure 5, calculate the error rate of the reference image in all areas, set the threshold K ₁ , when the error rate of the reference image When , it is determined as a region of interest (Region of Interest, ROI); when When , it is determined to be an uninteresting region; the selection of the threshold K ₁ depends on the requirements of the matching algorithm, and the commonly used thresholds are 0.4, 0.5, and 0.6.

2. Interpolation error rate

Suppose the total number of data holes in REM is M, and the number of missing data in the mth data hole is N _m ; a total of Q regions are determined as ROIs, and the ROI whose center coordinates are (i _q , j _q ) is denoted as REM mth data hole with The hole interpolation error rate between It is defined as follows:

in: is the missing elevation value of the nth data in the mth data hole of the REM after interpolation processing; for The elevation value relative to the nth data missing location of the mth data hole in the REM.

REM and The interpolation error rate between It is defined as follows:

Where max means to take the maximum value. The interpolation error rate ratio _ROI is defined as follows:

in: Indicates REM with The interpolation error rate between; min means take the minimum value. Set the threshold K ₂ , when the data hole area ratio _ROI ≤ K ₂ , it is judged as a small hole area; when the data hole area ratio _ROI > K ₂ , it is judged as a large hole area; the selection of the threshold K ₂ depends on the region’s The tolerance of the data hole size, commonly used thresholds are 0.10, 0.15, and 0.20.

3. Adaptive interpolation/segmentation scheme

Aiming at different sizes of data holes in REM, the data holes are classified by using the reference map error rate and interpolation error rate, and corresponding processing methods are proposed for data holes of different sizes, among which: for smaller data holes, the interpolation method is used to Fill in the missing elevation data; for larger data holes, adopt cutting processing, the cutting method is shown in Figure 6, if the largest image size REM after cutting meets the minimum template requirements, continue to match, otherwise, give up matching, commonly used The template values are 60×60, 80×80, 100×100, 120×120, etc.

The advantage of the present invention compared with prior art is:

(1) The present invention fully considers the data void problem faced by three-dimensional terrain matching in practical applications, and proposes a solution to improve the performance of terrain-assisted navigation.

(2) The present invention adopts the threshold discrimination method of the reference image error rate and the interpolation error rate, which is simple to calculate, high in reliability, and easy to realize in engineering.

Description of drawings

Fig. 1 is the structural representation of terrain aided navigation method;

Fig. 2 is a schematic diagram of acquisition of regions to be matched in real-time matching;

FIG. 3 is a schematic diagram of the principle of data hole generation;

Figure 4 is a schematic diagram of data holes appearing in the REM;

Fig. 5 is a schematic diagram of the traversal during the calculation of the error rate of the reference map;

FIG. 6 is a schematic diagram of data hole cutting;

Fig. 7 is a flowchart of a data hole adaptive interpolation/segmentation scheme.

Detailed ways

Aiming at the data hole problem existing in the three-dimensional terrain elevation matching, the present invention proposes a data hole self-adaptive interpolation/segmentation method. In order to make the implementation objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings:

In the data hole self-adaptive interpolation/segmentation method proposed by the present invention, data holes are divided into small data holes and large data holes, and bilinear interpolation is used for small data holes to complete missing data processing; large data holes are segmented , if the remaining area is not smaller than the minimum template size, continue to match, otherwise give up the match. The specific implementation process is shown in Figure 7 and can be described as follows:

(A) Obtaining REM through a three-dimensional elevation measurement sensor;

(B) With the current INS current position (i, j) as the center, determine a window in the reference terrain database as the area to be matched;

(C) Carry out data hole detection to REM, if REM contains data hole, then go to step (D); If REM has no data hole, then go to step (G);

(D) Calculate the reference image error rate between REM and DEM ^(i,j) when , it is judged as ROI; when , it is determined to be an area of no interest, and all ROI position information is stored; if there is an ROI area, then go to step (E), otherwise go to step (H);

(E) Process the REM with the interpolation method, and calculate the interpolation error rate ratio _ROI between the REM and the ROI. When the ratio _ROI ≤ K ₂ , it is judged as a small cavity area, and then go to step (G); when the ratio _ROI > K ₂ , it is determined to be a large cavity area, and then proceed to step (F);

(F) Carry out adaptive segmentation processing to REM, if the maximum frame REM after the segmentation processing meets the requirement of matching the minimum template, then proceed to step (D); otherwise, proceed to step (H);

(G) Utilize the three-dimensional elevation matching algorithm to match, correct the inertial navigation position according to the obtained matching position, and turn to step (1);

(H) Give up this match.

(1) read new REM, repeat above-mentioned operation, until matching ends.

The above embodiments are provided only for the purpose of describing the present invention, not to limit the scope of the present invention. The scope of the present invention is defined by the appended claims, and various equivalent replacements and modifications made without departing from the spirit and principle of the present invention shall fall within the scope of the present invention.

Claims (10)

1. the present invention proposes a kind of adaptive-interpolation/division processing method of data void holes, this method can be used for containing data sky The three-dimensional elevation of the REM in hole matches, characterized by comprising:

A REM) is obtained by three-dimensional elevation measurement sensor；

B) centered on the current location (i, j) of INS, in reference map database determine a search window, as to With region, region to be matched includes I × J DEM in total, and centre coordinate is that the DEM of (i, j) is denoted as DEM^(i,j)；

C data void holes detection) is carried out to REM and is transferred to step D if REM contains data void holes)；Otherwise, it is transferred to step G)；

D REM and DEM) is calculated^(i,j)Between reference map error rateWhenWhen, then it is determined as ROI；WhenWhen, then it is determined as region of loseing interest in.Region to be matched is traversed, all reference map error rates are calculated, storage is all ROI location information.ROI region if it exists is then transferred to step E)；Otherwise, it is transferred to step H)；

E) REM is handled using interpolation method, calculates the interpolation error rate ratio between REM and all ROI_ROI, work as ratio_ROI ≤K₂When, it is judged as small hole region, is transferred to step G)；Work as ratio_ROI> K₂When, it is determined as macroscopic-void region, is transferred to step F)；

F adaptivenon-uniform sampling processing) is carried out to REM, if the maximum map sheet REM after dividing processing, which meets, matches minimum template requirement, Then it is transferred to step D)；Otherwise, it is transferred to step H)；

G it) is matched using three-dimensional elevation matching algorithm, the inertial navigation position is corrected according to obtained matching position；

H this matching) is abandoned.

2. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step A) used in measurement of higher degree sensor further comprise InSAR, LiDAR, stereo vision camera, ultrasonic distance measurement Instrument, multibeam sonar and infrared ambulator etc..

3. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step B) further comprise:

Centered on the current location (i, j) of INS, according to INS longitude and latitude direction position error estimate the larger value σ determines region to be matched by 3 σ criterion in reference map database.

4. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step C) further comprise:

The presence or absence of data void holes in REM is judged using connected domain detection, and data void holes region is marked, wherein REM number Sum according to normal point is P, and the number of data void holes is M, and the shortage of data number in than the m-th data cavity is N_m。

5. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step D) further comprise:

Reference map error rateIt is defined as follows:

Wherein:Indicate REM and DEM_(i,j)Between reference map error rate, the normal sum of data is P in REM；f_REM (p) the REM height value normally located for p-th of data；f_DEM(p) DEM normally located for p-th of data corresponding with REM^(i,j)It is high Journey value.

6. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step E) interpolation method further comprises that linear interpolation method, bilinear interpolation, Kriging regression method, nearest neighbor point are inserted Value method, Natural neighbors interpolation method, minimum-curvature method, image factoring, radial basis function method and inverse distance multiply method etc..

7. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step E) the interpolation error rate further comprises:

Using reference map error rate, if shared Q region is judged as ROI, centre coordinate is (i_q,j_q) ROI be denoted asInterpolation error rate ratio_ROIIt is defined as follows:

Wherein:Indicate REM withBetween interpolation error rate；Min expression is minimized.

8. data void holes adaptive-interpolation/division processing method according to claim 1, it is characterised in that:

Step F) further comprise:

Adaptivenon-uniform sampling processing utilizes interpolation error rate as claimed in claim 7, and REM is cut in label maximum data cavity Cut processing.Whether it is not less than minimum stencil value according to the REM map sheet after cutting, judges whether to match, common minimum modulus Plate value has 60 × 60,80 × 80,100 × 100,120 × 120 etc..

9. interpolation error rate according to claim 7, it is characterised in that:

The REM withBetween interpolation error rate further comprise:

According to the number of REM data void holes known to claim 4 be M, REM withBetween interpolation error rateIt is fixed Justice is as follows:

Wherein:Indicate REM than the m-th data cavity withBetween empty interpolation error rate；Max expression takes maximum Value.

10. interpolation error rate according to claim 9, it is characterised in that:

The cavity interpolation error rate further comprises:

It is N according to the shortage of data number in the than the m-th data cavity of REM known to claim 4_m, REM than the m-th data cavity withBetween empty interpolation error rateIt is defined as follows:

Wherein:For the height value of the nth data missing in the than the m-th data cavity of REM after interpolation processing；ForThe height value of the nth data deletion sites in the than the m-th data cavity relative to REM.