CN106204547B - The method that rod-shaped atural object spatial position is automatically extracted from Vehicle-borne Laser Scanning point cloud - Google Patents

Abstract

The invention discloses a method for automatically extracting the spatial position of rod-shaped ground objects from the vehicle-mounted laser scanning point cloud. First, the optimal spatial layered point cloud plane projection image is obtained from the vehicle-mounted laser scanning point cloud; Threshold segmentation is performed on the spatially layered point cloud planar projection image to remove points with low brightness; line detection is performed on the threshold-segmented planar projection image to remove data with line features; further image extraction is performed to remove rod-shaped features In the data part of the diameter feature, the projected image of the rod-shaped object is obtained; finally, from the projected image of the rod-shaped object, the geometric center of each rod-shaped object area is taken as the spatial position positioning point of the rod-shaped object, and its The relative position is restored to the 3D point cloud. The method of the invention is not easily affected by data noise points, has a high degree of automation, fully utilizes the morphological characteristics of the point cloud data to a greater extent, and achieves a better extraction effect.

Description

Method for Automatically Extracting the Spatial Position of Rod-shaped Objects from Vehicle-mounted Laser Scanning Point Cloud

technical field

The invention belongs to the technical field of vehicle-mounted laser scanning point cloud data processing.

Background technique

As an advanced measurement method, the vehicle-mounted mobile laser measurement system is more and more widely used in urban 3D data collection. The 3D information collected by the system includes buildings, trees, light poles, power lines, bridges and roads on both sides of the road. pavement etc. Rod-shaped objects are the most common facilities in urban components. With the rapid development of smart cities, it is urgent to obtain more comprehensive and accurate spatial location information of rod-shaped objects. At present, for the research on the extraction of rod-shaped objects in laser point cloud, there are mainly clustering method and projection density method. Both methods are based on the overall point cloud data and are easily affected by noise points in the point cloud data. Sex is not high. How to better mine the morphological characteristics of point cloud data and improve the accuracy and efficiency of spatial position extraction of rod-shaped ground objects is still one of the current research difficulties.

Contents of the invention

In view of the above-mentioned technical problems, the present invention proposes a vehicle-mounted laser scanning point cloud based on the point-based projection image generated by studying the point-vehicle laser scanning data, using image processing, and combining the geometry and characteristics of rod-shaped features. The method for automatically extracting the spatial position information of rod-shaped objects, extracting the rod-shaped objects, can quickly and automatically extract the spatial position information of rod-shaped objects from massive point cloud data.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for automatically extracting the spatial position of a rod-shaped object from a vehicle-mounted laser scanning point cloud, comprising the following steps:

The first step is to obtain the spatially layered point cloud data of the rod-shaped object in the vertical direction in the local range from the vehicle-mounted laser scanning point cloud, and planarly project the spatially layered point cloud. Range and plane coordinates, project the discrete laser point cloud onto the plane, define the brightness value of the pixel with the number of point clouds in the unit pixel, convert it into two-dimensional image data, and automatically obtain the optimal spatial layered point cloud plane projection image ;

The second step is to perform threshold segmentation on the generated optimal spatial layered point cloud planar projection image, and remove points with low brightness, that is, to remove a small number of laser points in the same height range in the point cloud data;

The third step is to perform line detection on the planar projection image after threshold segmentation, and remove the data with line features, that is, remove the wall data of the building in the point cloud data;

The fourth step is to further extract the image, remove the data part that does not meet the diameter characteristics of the rod-shaped object, and obtain the projection image of the rod-shaped object;

The fifth step is to take the geometric center of each rod-shaped object area from the projected image of the rod-shaped object as the spatial position positioning point of the rod-shaped object, and restore its relative position to the 3D point cloud.

The advantages of the present invention are:

According to the laser point cloud data, the present invention automatically acquires the spatial layered data, and automatically selects the optimal spatial layered data for plane projection, and analyzes the characteristics of the rod-shaped objects to perform the rod-shaped objects on the point-based planar projection image. Extraction is not easily affected by data noise points, has a high degree of automation, and makes full use of the morphological characteristics of point cloud data to a greater extent, achieving better extraction results.

Description of drawings

Fig. 1 is the flowchart that the present invention implements;

Figure 2 is the automatic acquisition of the optimal spatial layered point cloud projection image;

Figure 3 is a layered point cloud projection image with many different high-level points;

Figure 4 is a layered point cloud projection image after filtering;

Figure 5 is the projection image of the spatial position of the rod-shaped object.

The projected images shown in Figures 2 to 5 are black-and-white-dot images obtained by inverting the black-and-white-dot images.

Detailed ways

Those skilled in the art can implement the present invention according to the content of the invention and the flow chart accompanying drawing 1 of implementation. For ease of implementation, each step in the summary of the invention will be described in further detail below, and the specific examples of Fig. 2-Fig. 5 are provided during the detailed description. Extract as an example.

1. Detailed description of the first step of the content of the invention:

1. Automatic extraction of layered point clouds in vertical space

According to the actual spatial height distribution of buildings, firstly, the point cloud is layered according to the vertical direction as a whole, and a point cloud layer is divided every 2m, and a series of spatial layered point cloud data in the vertical direction in the local range are obtained, which is set as l ₁ , l ₂ , l ₃ ... l _N ;

2. Plane projection of each point cloud layer in turn

Taking the xy plane as the projection plane, and the negative direction of the z direction as the projection direction, the laser point cloud is projected, and buildings, roads, rod-shaped features, etc. are projected into the projected coordinate system;

3. Generate a point-based planar projection image

According to the current point cloud range, set the point cloud coordinates as (X, Y, Z), the point cloud range as {X _min , Y _min , Z _min , X _max , Y _max , Z _max }, and the pixel coordinates of the image as (x , y), the scaling precision is s, then the point cloud coordinates (X, Y, Z) corresponding to the image (x, y) are: x=(XX _min )×s, y=(YY _min )×s;

The brightness of a point in a unit pixel is the superposition of the brightness of the number of point clouds falling in the unit pixel. Assuming that each point cloud is converted into (x, y), the brightness value of the point increases by 30%;

Traverse all point cloud data, calculate its corresponding plane projection image coordinates, and perform brightness superposition to obtain point-based plane projection images of all point cloud layers in the optimal space, thereby generating N plane projection images:

4. Automatically obtain the optimal spatial layered point cloud

Roughly detect the circumscribed circle of the block-shaped area on the generated N planar projection images, calculate the circumscribed circle diameter, and select the planar projection image with more diameters close to the actual rod-shaped object diameter as the optimal spatial layered point cloud projection image , proceed to the next step of rod-shaped feature extraction.

Figure 2 shows one of the optimal spatially layered point cloud projection images, which contains multiple pole-shaped feature point cloud projection images, and the pole-shaped feature pixels closest to the upper building are P1 and P2.

In order to prove the accuracy of the present invention, it is known that P1 and P2 are projected into the projected coordinate system, and the coordinates of P1 (509170.2306, 3985536.5028, 75.0464), and the coordinates of P2 (509173.3905, 3985537.3584, 74.6881), after being converted into a plane projection image, the coordinates are P1 'The coordinates are (293,161), and the coordinates of P2' are (354,181).

2. Detailed description of the second step of the content of the invention:

Use the existing adaptive threshold method to perform threshold segmentation on the optimal spatial layered point cloud projection image shown in Figure 2, that is, calculate the weighted average of the area around the pixel point, and then subtract a constant to obtain the adaptive threshold. Threshold segmentation is performed on the obtained planar projection image to obtain areas with higher brightness.

As shown in Figure 3, the positions of P1' and P2' have relatively high brightness, so they are well preserved. This part of the area corresponds to the data of many different high-level points within the unit range perpendicular to the xy plane in the point cloud, which is consistent with the feature that there are many points in the vertical direction of the rod-shaped object.

Three, a detailed description of the third step of the content of the invention:

Line detection is performed on the image after threshold segmentation, the parts with obvious wired characteristics in the image are removed, and these parts are set as the background color. The influence of the building walls in the actual 3D point cloud data on the extraction results of rod-shaped features.

As shown in Figure 4, after the image shown in Figure 3 is excluded from the influence of the building walls, the image shown in Figure 4 can be obtained.

Fourth, the detailed description of the fourth step of the content of the invention:

Local region growth is performed on the part with high brightness in the image after the influence of the building wall is excluded, and each region with high brightness is extracted, and these regions are filtered: first extract the boundary of a single region, if the circularity of the boundary of the region is less than 0.2, the area is linear and does not conform to the characteristics of rod-shaped objects, so exclude it; then extract a single area, calculate the diameter of the smallest circumscribed circle, and convert it into the distance in the point cloud data. If the data is much smaller or larger than If the diameter of the actual rod-shaped feature does not conform to the characteristics of the rod-shaped feature, it is excluded; and then judge whether the distance of n adjacent rod-shaped features conforms to the actual distance of the rod-shaped feature, if it matches, keep it, otherwise delete it.

As shown in FIG. 5 , after filtering the image shown in FIG. 4 , the projected image of the pole-shaped object shown in FIG. 5 is finally obtained.

Fifth, a detailed description of the fifth step of the content of the invention:

1. Calculate the center point of the rod-shaped object

For the remaining area of the projected image of the rod-shaped feature, extract the boundary of the area, and find the center of the smallest circumscribed circle as the center point of the rod-shaped feature;

2. Convert the coordinates of the center point of the rod-shaped object into point cloud coordinates

If the point coordinates of the rod-shaped object are set to (x, y), the corresponding point cloud coordinates (X, Y, Z) are: X=x/s+X _min , Y=y/s+Y _min , and the coordinates in the Z direction Take the intermediate value in the Z direction when the point cloud is spatially layered; the obtained (X, Y, Z) is the spatial position of the rod-shaped object.

As shown in Figure 5, the coordinates P1'(293,162) and P2'(354,179) of the center points of P1 and P2 in the image shown in Figure 4 are converted into point cloud coordinates as follows:

P1'(509170.2535, 3985536.5059, 75.0235);

P2' (509173.2149, 3985537.4218, 75.0235).

The point cloud coordinates converted above are basically consistent with the coordinates projected into the projected coordinate system, which proves that the method of the present invention can be applied to the extraction of actual rod-shaped features.

Claims (6)

1. A method for automatically extracting the spatial position of rod-shaped features from the vehicle-mounted laser scanning point cloud, characterized in that, comprising the steps: The first step is to obtain the spatially layered point cloud data of the rod-shaped object in the vertical direction in the local range from the vehicle-mounted laser scanning point cloud, and planarly project the spatially layered point cloud. Range and plane coordinates, project the discrete laser point cloud onto the plane, define the brightness value of the pixel with the number of point clouds in the unit pixel, convert it into two-dimensional image data, and automatically obtain the optimal spatial layered point cloud plane projection image ; The second step is to perform threshold segmentation on the generated optimal spatial layered point cloud planar projection image, and remove points with low brightness, that is, to remove a small number of laser points in the same height range in the point cloud data; The third step is to perform line detection on the planar projection image after threshold segmentation, and remove the data with line features, that is, remove the wall data of the building in the point cloud data; The fourth step is to further extract the image, remove the data part that does not meet the diameter characteristics of the rod-shaped object, and obtain the projection image of the rod-shaped object; The fifth step is to take the geometric center of each rod-shaped object area from the projected image of the rod-shaped object as the spatial position positioning point of the rod-shaped object, and restore its relative position to the 3D point cloud. 2. the method for claim 1 is characterized in that, the detailed steps of described first step are: 1.1. Automatic extraction of vertical spatial layered point cloud According to the actual spatial height distribution of buildings, firstly, the point cloud is layered according to the vertical direction as a whole, and a point cloud layer is divided every 2 meters, and a series of spatially layered point cloud data in the vertical direction in the local range are obtained, which is set to l ₁ , l ₂ , l ₃ ... l _N ; 1.2. Plane projection of each point cloud layer in turn Take the xy plane as the projection plane, and the negative direction of the z direction as the projection direction, project the laser point cloud, and project the buildings, roads, and pole-shaped objects into the projected coordinate system; 1.3. Generate a point-based planar projection image According to the current point cloud range, set the point cloud coordinates as (X, Y, Z), the point cloud range as {X _min , Y _min , Z _min , X _max , Y _max , Z _max }, and the pixel coordinates of the image as (x , y), and the scaling accuracy is s, then the point cloud coordinates (X, Y, Z) corresponding to the image (x, y) are: x=(XX _min )×s, y=(YY _min )×s; The brightness of a point in a unit pixel is the brightness superposition of the number of point clouds falling in the unit pixel, assuming that each point cloud is converted into a pixel point (x, y), the brightness value of the pixel point increases by 30%; Traverse all point cloud data, calculate its corresponding plane projection image coordinates, and perform brightness superposition to obtain point-based plane projection images of all point cloud layers in the optimal space, thereby generating N plane projection images; 1.4. Automatically obtain the optimal spatial layered point cloud Roughly detect the circumscribed circle of the block-shaped area on the generated N planar projection images, calculate the circumscribed circle diameter, and select the planar projection image with more circumscribed circle diameters close to the diameter of the actual rod-shaped object as the optimal spatial layered point cloud Project the image for the next step of rod-shaped feature extraction. 3. the method for claim 1 is characterized in that, the detailed steps of described second step are: In threshold segmentation, the adaptive threshold method is used to segment the optimal spatial layered point cloud projection image, that is, the weighted average of the area around the pixel is calculated, and then a constant is subtracted to obtain the adaptive threshold, and the obtained plane Threshold segmentation is performed on the projected image to obtain areas with higher brightness; this area corresponds to the data of many different high-level points in the unit range perpendicular to the xy plane in the point cloud, which is consistent with the fact that there are more points in the vertical direction of rod-shaped objects Characteristics. 4. method as claimed in claim 1 is characterized in that, the detailed steps of described 3rd step are: Perform line detection on the image after threshold segmentation, remove the parts with obvious wired characteristics in the image, and set these parts as the background color, so that the influence of the building walls in the actual 3D point cloud data on the extraction results of rod-shaped features can be eliminated . 5. the method for claim 1 is characterized in that, the detailed steps of described 4th step are: Perform local region growth on the part with high brightness in the image, extract each region with high brightness, and filter these regions: first extract the boundary of a single block, if the circularity of the boundary of the region is <0.2, the region is linear , does not conform to the characteristics of rod-shaped objects, and excludes them; then extracts a single area, calculates the diameter of the smallest circumscribed circle, and converts it into the distance in the point cloud data. If the distance is much smaller or larger than the actual rod-shaped object diameter, then If it does not conform to the characteristics of the rod-shaped object, it is excluded; the projection image of the rod-shaped object is obtained. 6. The method according to claim 5, characterized in that, the detailed steps of the fifth step are: 5.1. Calculating the center point of rod-shaped objects For the remaining area of the projection image of the rod-shaped object, extract the boundary of the single area, and find the center of the smallest circumscribed circle as the center point of the rod-shaped object; 5.2. The coordinates of the center point of the rod-shaped object are converted into point cloud coordinates The coordinates of the center point of the rod-shaped object are set to (x, y), and the corresponding point cloud coordinates (X, Y, Z) are: X=x/s+X _min , Y=y/s+Y _min , Z direction Take the intermediate value in the Z direction when spatially layering the point cloud; The calculated (X, Y, Z) is the spatial position of the bar-shaped object.