Unmanned aerial vehicle obstacle detection method in high-dust environment

Abstract

The invention provides an unmanned aerial vehicle obstacle detection method in a high-dust environment. A traditional unmanned aerial vehicle obstacle detection technology has many limitations in a high-dust environment, and the method aims at solving the problem that three-dimensional point cloud data obtained by a multi-line laser radar is combined with intensity information and curvature characteristics to recognize obstacles so as to guarantee the flight safety and stability of an unmanned aerial vehicle in the high-dust environment. The unmanned aerial vehicle has the advantages that high-precision, long-distance and large-range obstacle detection of the unmanned aerial vehicle in a high-dust environment is realized, and the unmanned aerial vehicle has the advantages of high stability and reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and particularly to a method for detecting obstacles of an unmanned aerial vehicle in a high-dust environment. Background Art

[0002] With the rapid development and application of unmanned aerial vehicle technology, its application scenarios in various environments have been continuously expanded, including but not limited to aerial photography, inspection, agriculture, fire fighting and rescue, etc. However, in some special environments, such as high-dust environments, traditional unmanned aerial vehicle obstacle detection technologies are severely restricted, which may lead to unsafe flight or abnormal operation of the unmanned aerial vehicle.

[0003] Traditional unmanned aerial vehicle obstacle detection technologies mainly include ultrasonic sensors, infrared sensors, and vision sensors, etc. Ultrasonic sensors are commonly used for short-distance obstacle detection and have good detection effects for some obstacles close to the unmanned aerial vehicle. However, in a high-dust environment, ultrasonic sensors are easily interfered by dust particles, resulting in limited ranging and inability to accurately detect long-distance obstacles. Infrared sensors are suitable for detecting short-distance obstacles and have the advantage of fast response, but are sensitive to the transparency and surface characteristics of obstacles and may fail or the detection accuracy may decrease in a high-dust environment. As a powerful environmental perception tool, vision sensors have broad application prospects in unmanned aerial vehicle obstacle avoidance. However, vision sensors have high requirements for lighting conditions and environmental backgrounds and may not work properly in a high-dust environment, thus affecting the effect of obstacle detection.

[0004] In contrast, lidar is one of the most widely used sensors in current unmanned aerial vehicle obstacle avoidance. Lidar has the advantages of high precision, long distance, and large range, and can accurately detect the positions and shapes of surrounding obstacles. However, in a high-dust environment, lidar may also be affected to a certain extent. For example, dust particles will scatter radar signals, resulting in reduced accuracy and reliability of obstacle detection.

[0005] Therefore, aiming at the limitations of traditional unmanned aerial vehicle obstacle detection technologies in a high-dust environment, there is an urgent need for a new method to improve the flight safety and stability of unmanned aerial vehicles in such a harsh environment. Summary of the Invention

[0006] The present invention provides a method for detecting obstacles of an unmanned aerial vehicle in a high-dust environment, which mainly includes the following steps:

[0007] S01: The unmanned aerial vehicle obtains the point cloud data of the lidar at a certain frequency, assuming the coordinates are and converts the coordinates to the body coordinate system

[0008]

[0009] Among them, the superscripts L and B respectively represent the radar coordinate system and the body coordinate system, the subscript N represents the total number of points in each frame of point cloud data, and R is the rotation matrix from the lidar coordinate system to the body system.

[0010] S02: Set the detection range, filter the points outside the detection range to obtain valid point cloud data

[0011]

[0012] Among them, k m is the index of the point (the serial number of the point in a frame of point cloud data), and M is the number of the valid point cloud.

[0013] S03: Determine whether the number M of the valid point cloud is less than the valid point number threshold TH OMN , if it is less than the valid point number threshold TH OMN then it is determined that there is no obstacle within the detection range, otherwise execute the following steps.

[0014] S04: According to the continuity of the points in the valid point cloud data, determine whether there is an obstacle within the detection range.

[0015] For ordered point cloud, count the point sets with continuous indexes of the valid point cloud data, and record the M c continuous points with the largest number of points; for unordered point cloud, use the k-means clustering method or the density-based DBSCAN clustering method to divide the valid point cloud data into multiple point sets, and record the M c continuous points with the largest number of points;

[0016] Determine whether the maximum number M of continuous points c is less than the continuous point number threshold TH OCN , if it is less than the continuous point number threshold TH OCN then it is determined that there is no obstacle within the detection range, otherwise execute the following steps.

[0017] S05: According to the smoothness of the points in the valid point cloud data, determine whether there is an obstacle within the detection range.

[0018] Calculate the curvature of the M c continuous points, and the calculation formula of the curvature value c(k i ) of the i-th point is as follows:

[0019]

[0020] Among them, is a point in the neighborhood point set of this point, and n represents the number of points in this point set. For ordered point clouds, the neighborhood point set takes n / 2 points forward and backward with the index k of the current point i as the center; for unordered point clouds, the neighborhood point set is screened by judging that the Euclidean distance from this point is less than a set threshold.

[0021] Count the M c number of consecutive points with curvature less than the curvature threshold TH curv , and calculate the proportion of points less than the curvature threshold TH curv ;

[0022] Judge whether the proportion is less than the proportion threshold TH ratio . If it is less than the proportion threshold TH ratio , it is determined that there are no obstacles in the detection range, otherwise it is determined that there are obstacles in the detection range.

[0023] Preferably, after performing step S02, points less than the threshold TH IMN can also be filtered according to the intensity value of the point cloud data to obtain the effective point cloud data.

[0024] By using the point cloud data obtained by lidar, the present invention realizes high-precision, long-distance, and large-range detection of obstacles in a high-dust environment, effectively improving the flight safety and stability of the drone in such an environment. Description of the Drawings

[0025] Figure 1 is a flowchart of a method for detecting obstacles of a drone in a high-dust environment provided by Embodiment 1 of the present invention;

[0026] Figure 2 is a flowchart of a method for detecting obstacles of a drone in a high-dust environment provided by Embodiment 2 of the present invention. Detailed Embodiments

[0027] The embodiments of the present invention are described as follows:

[0028] Embodiment 1 is an outdoor park monitoring scenario. The drone is equipped with a solid-state lidar fixed and placed obliquely downward to ensure that there are no obstacles in the park affecting the activities of personnel and the operation of facilities. As Figure 1 shown, it is a flowchart of a method for detecting obstacles of a drone in a high-dust environment provided by Embodiment 1 of the present invention, including the following steps:

[0029] S101: Acquisition of point cloud data. The drone acquires the point cloud data of the solid-state lidar at a scanning frequency of 5 Hz and converts it to the body coordinate system, where the rotation matrix R is a fixed value measured in advance and known.

[0030] S102: Set the detection range and threshold parameters. Considering the open characteristics of the outdoor park, the set detection range is the full coverage range of the solid-state lidar's perspective to provide richer spatial information. The recommended range for setting the threshold parameters is as follows: the effective point number threshold TH OMN is 25 - 30, the continuous threshold TH OCN is 15 - 20, the curvature threshold TH curv is 0.4 - 0.6, and the ratio threshold TH ratio is 0.4 - 0.7.

[0031] S103: Determine the valid point cloud. First, judge whether the number of points in the valid point cloud data obtained within the set detection range exceeds the set effective point number threshold TH OMN , if it is less than TH OMN then it is determined that there are no obstacles within the detection range, otherwise continue with the following steps.

[0032] S104: Determine the continuity of points. Next, count the M c points with the largest number of points in the valid point cloud data, and judge whether M c is less than the set continuous point number threshold TH OCN , if it is less than TH OCN then it is determined that there are no obstacles within the detection range, otherwise continue with the following steps.

[0033] S105: Determine the smoothness of points. For those that meet the continuity requirements, further calculate the curvature of each point and count the proportion of points with a curvature less than the set curvature threshold TH curv , and judge whether this proportion is less than the set ratio threshold TH ratio , if it is less than TH ratio then it is determined that there are no obstacles within the detection range, otherwise it is determined that there are obstacles within the detection range.

[0034] In this embodiment, first, through steps S103 and S104, physical obstacles composed of continuous point cloud data, such as trees or buildings, can be effectively identified; then, through step S105, it is further determined whether the shape and smoothness of the obstacles meet the requirements. The final monitoring report will indicate whether there are major obstacles such as trees, buildings, and road obstacles in the park and provide the location information of the obstacles. According to the monitoring results, the flight path of the drone can be adjusted to ensure the safety of personnel activities and facility operations in the park.

[0035] Embodiment 2 is for the inspection scenario of an underground parking lot. The drone is equipped with a multi-line mechanical lidar placed horizontally and fixedly to ensure that there are no obstacles in the parking lot affecting vehicle entry and exit and facility operation. As Figure 2As shown in the figure, it is a flowchart of an obstacle detection method for an unmanned aerial vehicle (UAV) in a high-dust environment provided by the second embodiment of the present invention, including the following steps:

[0036] S201: Acquisition of point cloud data. The UAV acquires the point cloud data of the multi-line lidar at a scanning frequency of 10 Hz and converts it into the body coordinate system, where the rotation matrix R is a fixed value measured in advance and known.

[0037] S202: Setting of detection range and threshold parameters. Considering the characteristics of few people and relatively fixed vehicle driving routes in the underground parking lot, the detection range is the range within 2 meters in front of the UAV within 90 degrees; the recommended setting range of the threshold parameters is as follows: the effective point number threshold TH OMN is 15 - 25, the continuous threshold TH OCN is 5 - 15, the intensity threshold TH IMN is 30 - 40, the curvature threshold TH curv is 0.3 - 0.7, and the ratio threshold TH ratio is 0.4 - 0.6.

[0038] S203: Judgment of point cloud intensity value. Considering that the dusty environment in the underground parking lot will affect the obstacle detection, the intensity value of the point cloud is also used for filtering, and only the points with the intensity value greater than or equal to TH IMN are retained.

[0039] S204: Judgment of valid point cloud. First, judge whether the number of points in the valid point cloud data obtained within the detection range exceeds the set effective point number threshold TH OMN . If it is less than TH OMN , it is determined that there are no obstacles within the detection range; otherwise, continue with the following steps.

[0040] S205: Judgment of point continuity. Then, count the M c consecutive points with the most points in the valid point cloud data, and judge whether M c is less than the set consecutive point number threshold TH OCN . If it is less than TH OCN , it is determined that there are no obstacles within the detection range; otherwise, continue with the following steps.

[0041] S206: Judgment of point smoothness. For those that meet the continuity requirements, further calculate the curvature of each point and count the proportion of points with the curvature less than the set curvature threshold TH curv . Judge whether it is less than the set ratio threshold TH ratio . If it is less than TH ratio , it is determined that there are no obstacles within the detection range; otherwise, it is determined that there are obstacles within the detection range.

[0042] In this embodiment, considering the relatively large amount of dust in the underground parking lot, the method uses intensity values for effective point cloud filtering, and then determines whether there are obstacles within the detection range based on the continuity and smoothness of the points. The final detection result will indicate whether there are obstacles such as walls, columns, parked vehicles, etc. in the underground parking lot, and ensure the safe driving and parking of the vehicle by controlling the flight path of the drone.

[0043] Embodiment 3 is a scenario of mine inspection. The drone is equipped with a rotary multi-line lidar to scan the obstacle situation in the mine omni-directionally. In this embodiment, the specific flowchart of the drone obstacle detection method is the same as that of Embodiment 2, including the following steps: S301: Acquisition of point cloud data. The drone acquires the point cloud data of the multi-line lidar at a scanning frequency of 20 Hz. Considering the narrow channels in the mine to avoid missed detection, the multi-line lidar rotates around the X-axis of the radar coordinate system at a rotational speed of 30 - 90 degrees per second, and then converts the point cloud data into the body coordinate system, where the rotation matrix R is obtained by real-time measurement of the IMU sensor fixedly connected to the multi-line lidar.

[0044] S302: Setting of detection range and threshold parameters. Considering the high risk of the mine, the detection range is the 360-degree omnidirectional range covered by the view angle of the multi-line lidar; the recommended setting range of the threshold parameters is as follows: the effective point number threshold TH OMN is 10 - 15, the continuous threshold TH OCN is 5 - 15, the intensity threshold TH IMN is 20 - 30, the curvature threshold TH curv is 0.4 - 0.8, and the ratio threshold TH ratio is 0.5 - 0.8.

[0045] S303: Judgment of point cloud intensity value. Considering the influence of the dusty environment underground on obstacle detection, the intensity value of the point cloud is also used for filtering, and only the points with intensity values greater than or equal to TH IMN are retained.

[0046] S304: Judgment of effective point cloud. First, judge whether the number of points in the acquired effective point cloud data within the detection range exceeds the set effective point number threshold TH OMN , if it is less than TH OMN , it is determined that there are no obstacles within the detection range, otherwise continue the following steps.

[0047] S305: Judgment of point continuity. Next, count the M c continuous points with the largest number of points in the effective point cloud data, and judge whether M c is less than the set continuous point number threshold TH OCN , if it is less than TH OCNIt is determined that there is no obstacle within the detection range; otherwise, continue with the following steps.

[0048] S306: Judgment of point smoothness. For those that meet the continuity requirement, further calculate the curvature of each point and count the proportion of points with curvature less than the set curvature threshold TH curv to determine whether it is less than the set proportion threshold TH ratio based on this proportion. If it is less than TH ratio it is determined that there is no obstacle within the detection range; otherwise, it is determined that there is an obstacle within the detection range.

[0049] In this embodiment, according to the characteristics of the dark mine environment and relatively large dust particles, a relatively low effective point number threshold TH OMN is set, as well as a continuous point number threshold TH OCN and an intensity threshold TH IMN , and a relatively loose curvature threshold TH curv and a proportion threshold TH ratio . These adjustments are to ensure that obstacles can be effectively detected even under poor visual conditions. Finally, the system generates an inspection report indicating whether there are major obstacles in the mine, such as mine cars, mine pillars, equipment pipelines, etc., and provides detailed obstacle location information. According to the inspection results, the flight path of the drone is adjusted to ensure the normal operation of mining equipment and the safety of miners.

[0050] The above is the preferred implementation manner of the present invention, which expounds the applicability and flexibility of the method under different environmental conditions. Especially in the case of requiring full - range scanning and precise obstacle positioning, it provides important support for the safe navigation and effective task execution of the drone. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle described in the present invention, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for detecting obstacles of unmanned aerial vehicles in a high dust environment, comprising the following steps: a. The drone obtains the point cloud data of the laser radar at a certain frequency and converts it to the body coordinate system; b. Set the detection range, filter the points outside the detection range, and obtain valid point cloud data; c. Determine whether the number of points in the valid point cloud data is less than the valid point threshold TH OMN If it is less than the effective point threshold TH OMN It is determined that there is no obstacle within the detection range, otherwise the detection continues; d. judging whether there are obstacles within the detection range according to the continuity of the midpoints of the valid point cloud data; e. Determine whether there are obstacles within the detection range based on the smoothness of the points in the valid point cloud data.

2. The method according to claim 1, characterized in that After setting the detection range, filtering points outside the detection range, and obtaining valid point cloud data, the method may further include: According to the intensity value of the point cloud data, the points with an intensity less than the threshold value TH are filtered out. IMN points to obtain the valid point cloud data.

3. The method according to claim 1, characterized in that The determining whether there is an obstacle within the detection range according to the continuity of the midpoints of the valid point cloud data includes: For an ordered point cloud, count the indexed continuous point sets of the valid point cloud data and record the M with the largest number of points. c consecutive points; or, For unordered point clouds, the valid point cloud data is divided into multiple point sets using a clustering method, and the M with the largest number of points is recorded. c continuous points, the clustering method includes a k-means clustering method or a density-based DBSCAN clustering method; Judge M c Is it less than the continuous point threshold TH? OCN If it is less than the continuous point threshold TH OCN It is determined that there is no obstacle within the detection range, otherwise the detection continues.

4. The method according to claim 1, characterized in that: The determining whether there is an obstacle within the detection range according to the smoothness of the midpoints of the valid point cloud data includes: Calculate the M c The curvature of consecutive points; Statistics of M c The curvature of the continuous points is less than the curvature threshold TH curv The number of points less than the curvature threshold TH is calculated curv The proportion of the number of points; Determine whether the ratio is less than the ratio threshold TH ratio , if it is less than the ratio threshold TH ratio It is determined that there is no obstacle within the detection range, otherwise it is determined that there is an obstacle within the detection range.

Images