A surface displacement monitoring method based on UAV lidar detection

Abstract

The present invention relates to the field of slope monitoring technology, and more particularly to a surface displacement monitoring method based on unmanned aerial vehicle (UAV) laser radar detection. This method utilizes a multi-point UAV positioning detection method, which reduces the stability requirements for the drone during detection compared to mobile detection, resulting in more accurate data. By analyzing feature data constructed from the coordinates of the slope surface, three-dimensional mapping is avoided, significantly saving computing power, reducing hardware and software costs, and facilitating implementation. Data is collected twice on the overlapping slope between two adjacent slope sections. By analyzing the differences between the two data sets, the quality of the data collection can be determined, further improving monitoring accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of slope monitoring, and in particular to a surface displacement monitoring method based on unmanned aerial vehicle (UAV) laser radar detection. Background Art

[0002] During the construction of highways, slope landslides are a major safety hazard, and long-term monitoring of the slopes is required to determine the progress of the project. In the absence of external interference, the slopes will experience varying degrees of displacement before landslides. The two common landslide modes are: Figure 1 In either case, a small amount of slope displacement will occur before a large-scale landslide. Therefore, if the slope displacement before the landslide can be monitored, the construction safety hazards can be effectively reduced.

[0003] At present, common slope displacement monitoring methods mostly use measuring tools such as inclinometers and liquid level meters. Their shortcomings are: the highway construction section is long and the slope is long. Monitoring the entire slope section requires a large number of measuring instruments and the layout of a stable and reliable communication network. This is costly and difficult to implement, making it difficult to promote.

[0004] Of course, in recent years, there have also been technical means of using drones to monitor slope displacement. However, the shortcomings are that most drones use a combination of lidar and cameras, and the collected data requires point cloud processing to produce three-dimensional composition. Every time the drone obtains data, the three-dimensional map needs to be updated, which requires huge computing power and high software usage costs. Summary of the Invention

[0005] The present invention discloses a surface displacement monitoring method based on UAV laser radar detection, which can realize UAV slope monitoring without using three-dimensional composition, has low use cost and wide application range.

[0006] It is achieved through such a technical solution, the specific method is as follows:

[0007] For the slope to be monitored, a number of aerial sampling coordinates are planned in combination with the LiDAR preset effective wide-angle detection;

[0008] Each aerial sampling coordinate corresponds to a slope segment, and there are overlapping slope segments of the same range between two adjacent slope segments;

[0009] Control the UAV to fly to each aerial sampling coordinate in sequence, complete the detection of the corresponding slope section at each aerial sampling coordinate, obtain the surface coordinate information of each corresponding slope section and construct the corresponding window data;

[0010] Determine whether all window data have abnormalities. If so, remove the abnormal window data.

[0011] Analyze the window data of several different time periods to determine whether the slope has shifted.

[0012] The advantages of this embodiment are that it adopts a multi-point positioning detection method for drones, which can reduce the requirements for drone body stability during detection compared to mobile detection, and the collected data is more accurate; by analyzing the characteristic data constructed by the coordinates of the slope surface, three-dimensional composition is avoided, computing power is significantly saved, hardware and software costs are reduced, and it is easy to implement; two data collections are completed on the overlapping slope section between two adjacent slope sections, and by analyzing the difference between the two data, the quality of data collection can be judged, further improving the accuracy of monitoring.

[0013] Furthermore, several aerial sampling coordinates are planned. The specific method is as follows:

[0014] Determine the optimal detection range of the UAV-mounted lidar;

[0015] The first aerial sampling coordinate is preset at the starting point of the slope to be monitored. The boundary conditions are that the UAV's wide-angle detection boundary can cover the starting point of the slope to be monitored and the distance between the slope at the detection wide-angle boundary and the lidar does not exceed the optimal detection distance. The optimization condition is to maximize the slope area covered by the detection range. The position of the current aerial sampling coordinate is optimized using a dynamic programming algorithm.

[0016] The detection boundary of the previous aerial sampling coordinate is obtained, so that the drone has a preset overlap segment with the detection boundary of the previous aerial sampling coordinate when the current aerial sampling coordinate is located. Within the boundary conditions, the dynamic programming algorithm is used to optimize the position of the current aerial sampling coordinate with the maximum slope area covered by the detection range as the optimization condition;

[0017] Repeat the progressive planning of several aerial sampling coordinate positions until the detection range completely covers the entire slope to be monitored.

[0018] The advantage of this embodiment is that it proposes a method for automatically finding and planning aerial sampling coordinates, which reduces the number of sampling times as much as possible while meeting the sampling quality requirements and saving detection time.

[0019] Furthermore, if there is an obstacle at any aerial sampling coordinate, the obstacle is modeled and the side of the obstacle away from the slope to be monitored is set as the restricted coordinate. Within the boundary conditions, the aerial sampling coordinates with the current obstacle are re-optimized with the preset overlap segment with the detection boundary of the previous aerial sampling coordinate and the maximization of the slope area covered by the detection range within the boundary conditions as the optimization conditions.

[0020] After re-optimizing the coordinates of the current aerial sampling with obstacles, re-optimize the coordinates of subsequent aerial sampling in sequence.

[0021] The advantage of this embodiment is that it takes into account the diversity of external situations and provides a second-choice method when the optimal sampling position does not support sampling, so that the entire solution has good adaptability.

[0022] Furthermore, the corresponding slope sections are detected, and the surface coordinate information of each corresponding slope section is obtained to construct the corresponding window data. The specific method is as follows:

[0023] Use real-time dynamic carrier phase differential positioning algorithm to achieve precise positioning of UAVs;

[0024] Control the drone to fly to each aerial sampling coordinate, adjust the drone's flight attitude so that the preset effective detection wide angle can cover the corresponding slope section, and obtain the three-dimensional coordinates of several sampling points on the surface of each slope section through the lidar;

[0025] The height information of the three-dimensional coordinates of several sampling points on the surface of each slope section is read, and a matrix is ​​constructed with the height information according to the arrangement order of the collected samples. The matrix corresponds to the window data.

[0026] The advantage of this embodiment is that the dynamic carrier phase differential positioning technology can achieve high-precision coordinate positioning, so that the drone can sample at the same coordinates as much as possible, further reducing the possibility of error.

[0027] Furthermore, the abnormal window data is removed. The specific method is as follows:

[0028] Determine the matrix position corresponding to the overlapping part of the data of two adjacent windows;

[0029] Compare the data of the overlapping positions of two adjacent matrices one by one. If the height difference of the same position in the two matrices is greater than the upper threshold, assign the position a value of 1; if the height difference of the same position in the two matrices is less than the lower threshold, assign the position a value of -1; if it is between the upper and lower thresholds, assign the position a value of 0. Use the assigned data to construct an anomaly judgment matrix;

[0030] Input the abnormal judgment matrix into the convolutional neural network model to determine the cause of the abnormal judgment matrix;

[0031] If the cause is caused by an interference signal or an abnormality in the aerial sampling coordinates detected by the UAV, the two adjacent window data used to construct the interference matrix will be eliminated; otherwise, the two adjacent window data used to construct the interference matrix will be retained.

[0032] The advantage of this embodiment is that it utilizes the overlapping parts of adjacent slope detections to screen abnormal data. When the gap in the overlapping parts is too large, it may be caused by a landslide or various errors. The differences in the overlapping parts can be used to construct an abnormal judgment matrix. The differences caused by different reasons will have unique feature manifestations in space. By training the convolutional neural network, different feature manifestations can be extracted to determine the causes of the differences and complete the abnormal data screening.

[0033] Furthermore, the convolutional neural network model is constructed as follows:

[0034] The UAV is equipped with a laser radar to obtain the surface height of the test target in a shaking and balanced state, and the abnormal matrix of the training UAV is constructed based on the surface height in the shaking and balanced state;

[0035] The UAV-mounted laser radar is used to obtain the surface height of the test target in the presence of various interference sources, and a training interference anomaly matrix is ​​constructed based on the surface height of the test target under interference and without interference.

[0036] Construct training data using drone anomaly matrix, interference anomaly matrix, and effective matrix;

[0037] Train the convolutional neural network model using the training data.

[0038] Furthermore, we analyze the window data of several different time periods to determine whether the slope has shifted. The specific method is as follows:

[0039] Get window data of two different time periods on the same slope section;

[0040] Subtract the window data of two different time periods;

[0041] The number of sampling point changes is counted. If the preset number is reached, the absolute value of the window data after subtraction is accumulated and then divided by the number of sampling points of the slope section to obtain the displacement judgment value. If the displacement judgment value is greater than the threshold, it is judged that the slope section has shifted.

[0042] The advantage of this embodiment is that the landslide displacement has the characteristic of overall movement, and judging whether a landslide has occurred by the average displacement of the displacement points can improve the accuracy of the judgment.

[0043] Furthermore, the weight is calculated by obtaining the average absolute value of the window data after subtraction as the denominator, and using the insulation value of each data bit after subtraction as the numerator to calculate the weight value of each data bit in the window data.

[0044] The advantage of this embodiment is that the weight can amplify the influence of the displacement, thereby increasing the sensitivity of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings of the present invention are described below.

[0046] Figure 1 These are two common landslide modes in background technology.

[0047] Figure 2 Schematic diagram of converting slope sections into window data matrix.

[0048] Figure 3 Schematic diagram of the overlapping part of the data in two adjacent windows.

[0049] Figure 4 Schematic diagram of the height matrix of the same window comparison time.

[0050] Figure 5 Schematic diagram of the monitoring process. DETAILED DESCRIPTION

[0051] The present invention will be further described below with reference to the accompanying drawings and examples.

[0052] A surface displacement monitoring method based on UAV lidar detection, such as Figure 5 The specific steps are as follows:

[0053] S1. Plan several spatial sampling coordinates:

[0054] S11. Determine the optimal detection distance d of the UAV according to the laser radar parameter table. best and wide-angle detection Connect the laser radar to any point on the edge of the slope to be monitored, so that the angle between the laser radar and the connecting line is less than α. v =d best .cosα calculates the maximum vertical distance between the UAV and the slope to be monitored, and thus determines the range that meets the conditions as the boundary condition. Within the boundary condition range, several backup points are randomly selected and the slope coverage range of each backup point is calculated as the adaptation value. The calculation formula is as follows:

[0055] l=d best .cosθ-l overflow ,θ≤α

[0056] Among them, θ is the actual detection wide angle of the backup point, l overflow To overflow the length of the slope section, the spare point with the widest coverage was selected as the first aerial sampling coordinate.

[0057] S12. Determine the overlapped area. Two adjacent aerial sampling coordinates need to have a partially overlapping detection area. Subtract the overlapped area from the end point of the previous spatial sampling coordinate to use as the starting boundary line for calculating the current spatial coordinate.

[0058] S13. Repeat the calculation process of step S11 based on the starting boundary line when calculating the current spatial coordinates to calculate the current aerial sampling coordinates.

[0059] S14. After calculating each aerial sampling coordinate, calculate the next aerial sampling coordinate forward until the entire slope to be monitored is completely covered.

[0060] In step S1, if there is an obstacle that interferes with detection at an aerial sampling point, the obstacle is modeled and the side of the obstacle away from the slope to be monitored is set as the restricted coordinate. Within the boundary conditions, the aerial sampling coordinates of the current obstacle are re-optimized to meet the preset overlap segment with the detection boundary of the previous aerial sampling coordinate and to maximize the slope area covered by the detection range within the boundary conditions.

[0061] S2. Build corresponding window data:

[0062] S21. Use real-time dynamic carrier phase differential positioning algorithm to achieve precise positioning of drones.

[0063] S22. Control the drone to fly to each aerial sampling coordinate and adjust the drone's flight attitude so that the preset effective detection wide angle can cover the corresponding slope section.

[0064] S23. Several sampling points are set at equal intervals on each slope section. The actual distance between the several sampling points on the surface of each slope section and the laser radar is obtained by laser radar. The spatial coordinates of each sampling point are calculated based on the actual coordinates of the laser radar and the actual distance.

[0065] S24, read the height information of the three-dimensional coordinates of several sampling points on the surface of each slope section, and construct a matrix with the height information according to the arrangement order of the collected samples. The matrix corresponds to the window data, as shown in 2.

[0066] S3. Eliminate abnormal window data:

[0067] S31, determine the matrix position corresponding to the overlapping part of the data of two adjacent windows, such as Figure 3 shown.

[0068] S32. Compare the data of the overlapping positions of two adjacent matrices one by one. If the height difference of the same position in the two matrices is greater than 0.5, assign the position a value of 1; if the height difference of the same position in the two matrices is less than -0.5, assign the position a value of -1; if the height difference is between -0.5 and 0.5, assign the position a value of 0. Construct an anomaly judgment matrix based on the assigned data; as follows:

[0069]

[0070] S33. Input the abnormal judgment matrix into the convolutional neural network model to determine the cause of the abnormal judgment matrix.

[0071] The method for constructing the convolutional neural network model is as follows:

[0072] S331, causing the UAV to carry a laser radar to obtain the surface height of the test target in a shaking and balanced state, and constructing a training UAV anomaly matrix based on the surface height in the shaking and balanced state;

[0073] S332, causing the UAV-mounted laser radar to obtain the surface height of the test target in the presence of various interference sources, and constructing a training interference anomaly matrix based on the surface height of the test target under interference and the surface height of the test target without interference;

[0074] S333, constructing training data using the drone anomaly matrix, interference anomaly matrix, and effective matrix;

[0075] S334. Train the convolutional neural network model using the training data.

[0076] S34. If the cause is caused by an interference signal or an abnormality in the aerial sampling coordinates detected by the UAV, the two adjacent window data used to construct the interference matrix are eliminated; otherwise, the two adjacent window data used to construct the interference matrix are retained.

[0077] In steps S33 and S34 of this embodiment, the convolutional neural network model indicates that the anomaly is caused by the drone's displacement. Human reasoning can also determine that the overlapped portion has shifted overall from the rear of the detection overlap. Since the two detections were not separated by a long time, the most likely cause of the overall error is prolonged oscillation of the drone's fuselage while detecting the latter half of the sampling points. Similar to the convolutional neural network model's conclusion, this window of data should be deleted and not used for comparison.

[0078] S4. Analyze the window data of several different time periods to determine whether the slope has shifted:

[0079] S41, obtain the window data of the same slope section in two time periods, such as Figure 4 shown.

[0080] S42. Subtract the window data of two different time periods and take the absolute value to obtain the change matrix as follows:

[0081]

[0082] S43. Since the number of change points is 15, which is greater than the preset value of 5, the displacement judgment value is calculated according to the formula:

[0083]

[0084] Where m is the number of rows in the change matrix, n is the number of columns in the change matrix, and P mn is the value of the mth row and nth column of the change matrix, ω mn is the weight of the value in the mth row and nth column of the change matrix;

[0085]

[0086] In this embodiment, after the specific value of JV is calculated, it is compared with the judgment threshold. If it is greater than the threshold, it is determined that displacement has occurred, otherwise it is determined that no displacement has occurred.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A surface displacement monitoring method based on UAV laser radar detection, characterized in that: The specific method is as follows: For the slope to be monitored, a number of aerial sampling coordinates are planned in combination with the LiDAR preset effective wide-angle detection; Each aerial sampling coordinate corresponds to a slope segment, and there are overlapping slope segments of the same range between two adjacent slope segments; Control the UAV to fly to each aerial sampling coordinate in sequence, complete the detection of the corresponding slope section at each aerial sampling coordinate, obtain the surface coordinate information of each corresponding slope section and construct the corresponding window data; Determine whether all window data have abnormalities. If so, remove the abnormal window data. Analyze window data from several different time periods to determine whether the slope has shifted; Complete the detection of the corresponding slope section, obtain the surface coordinate information of each corresponding slope section and construct the corresponding window data. The specific method is as follows: Use real-time dynamic carrier phase differential positioning algorithm to achieve precise positioning of UAVs; Control the drone to fly to each aerial sampling coordinate, adjust the drone's flight attitude so that the preset effective detection wide angle can cover the corresponding slope section, and obtain the three-dimensional coordinates of several sampling points on the surface of each slope section through the lidar; Read the height information of the three-dimensional coordinates of several sampling points on the surface of each slope section, and construct a matrix with the height information according to the arrangement order of the collected samples. The matrix corresponds to the window data. Eliminate abnormal window data. The specific method is as follows: Determine the matrix position corresponding to the overlapping part of the data of two adjacent windows; Compare the data of the overlapping positions of two adjacent matrices one by one. If the height difference of the same position in the two matrices is greater than the upper threshold, assign the position a value of 1; if the height difference of the same position in the two matrices is less than the lower threshold, assign the position a value of -1; if it is between the upper and lower thresholds, assign the position a value of 0. Use the assigned data to construct an anomaly judgment matrix; Input the abnormal judgment matrix into the convolutional neural network model to determine the cause of the abnormal judgment matrix; If the cause is caused by interference signals or abnormal coordinates of aerial sampling detected by UAVs, the two adjacent windows of data used to construct the interference matrix are eliminated; Otherwise, the two adjacent window data used to construct the interference matrix are retained.

2. The surface displacement monitoring method based on UAV laser radar detection according to claim 1, characterized in that: Plan several aerial sampling coordinates. The specific method is as follows: Determine the optimal detection range of the UAV-mounted lidar; The first aerial sampling coordinate is preset at the starting point of the slope to be monitored. The boundary conditions are that the UAV's wide-angle detection boundary can cover the starting point of the slope to be monitored and the distance between the slope at the detection wide-angle boundary and the lidar does not exceed the optimal detection distance. The optimization condition is to maximize the slope area covered by the detection range. The position of the current aerial sampling coordinate is optimized using a dynamic programming algorithm. The detection boundary of the previous aerial sampling coordinate is obtained, so that the drone has a preset overlap segment with the detection boundary of the previous aerial sampling coordinate when the current aerial sampling coordinate is located. Within the boundary conditions, the dynamic programming algorithm is used to optimize the position of the current aerial sampling coordinate with the maximum slope area covered by the detection range as the optimization condition; Repeat the progressive planning of several aerial sampling coordinate positions until the detection range completely covers the entire slope to be monitored.

3. The surface displacement monitoring method based on UAV laser radar detection according to claim 2, characterized in that: If there is an obstacle at any aerial sampling coordinate, the obstacle is modeled and the side of the obstacle away from the slope to be monitored is set as the restricted coordinate. Within the boundary conditions, the aerial sampling coordinates with the current obstacle are re-optimized with the preset overlap segment with the detection boundary of the previous aerial sampling coordinate and the maximization of the slope area covered by the detection range within the boundary conditions as the optimization conditions; After re-optimizing the coordinates of the current aerial sampling with obstacles, re-optimize the coordinates of subsequent aerial sampling in sequence.

4. The surface displacement monitoring method based on UAV laser radar detection according to claim 1, characterized in that: The method for constructing the convolutional neural network model is as follows: The UAV is equipped with a laser radar to obtain the surface height of the test target in a shaking and balanced state, and the abnormal matrix of the training UAV is constructed based on the surface height in the shaking and balanced state; The UAV-mounted laser radar is used to obtain the surface height of the test target in the presence of various interference sources, and a training interference anomaly matrix is ​​constructed based on the surface height of the test target under interference and without interference. Construct training data using drone anomaly matrix, interference anomaly matrix, and effective matrix; Train the convolutional neural network model using the training data.

5. The surface displacement monitoring method based on UAV laser radar detection according to claim 1, characterized in that: Analyze the window data of several different time periods to determine whether the slope has shifted. The specific method is as follows: Get window data of two different time periods on the same slope section; Subtract the window data of two different time periods; The number of sampling point changes is counted. If the preset number is reached, the absolute value of the window data after subtraction is accumulated and then divided by the number of sampling points of the slope section to obtain the displacement judgment value. If the displacement judgment value is greater than the threshold, it is judged that the slope section has shifted.

6. The surface displacement monitoring method based on UAV laser radar detection according to claim 5, characterized in that: The weight value is calculated by taking the average absolute value of the window data after subtraction as the denominator, and taking the insulation value of each data bit after subtraction as the numerator, and calculating the weight value of each data bit in the window data.

Images