Tower crane collision detection method based on multi-source perception

Abstract

The invention discloses a tower crane collision detection method based on multi-source perception, the multi-source perception comprises a visible light camera and a depth camera, the visible light camera collects image data of an object, and the depth camera collects depth information of the object; processing the acquired image data and depth information to form the outline of the object, and calculating the outline proportion of the lifting hook and the lifted object; a pixel equivalent is established according to the imaging size of the lifting hook, and the true size of an unknown lifting object is calculated by directly applying the proportionality coefficient, so that the three-dimensional size of the lifting object is output; in the working process of the tower crane, constructing a mixed bounding box by utilizing the spatial pose information; a parallel collision detection algorithm based on the separation axis theorem is deployed, each surface normal of each bounding box provides a candidate separation axis, and two-stage collision threshold values are set through full-scene bounding box spatial topological relation analysis: the collision early warning problem in tower crane operation is effectively solved, and the tower crane operation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tower crane construction, and particularly relates to a tower crane collision detection method based on multi-source perception. Background Art

[0002] A tower crane is one of the most commonly used lifting equipment on construction sites, also known as a "tower hoist", which is composed of standard sections connected end to end, and is used to lift construction raw materials such as steel bars, wooden beams, concrete, and steel pipes for construction. Tower cranes are essential lifting equipment on construction sites. The operation of existing tower cranes is jointly controlled by a driver sitting in the tower crane cab and a ground signalman. The movement of the jib and trolley drives the movement of the hook. Since the driver and signalman are far from the hook when hoisting materials, there is a certain error in judging the position of the hook, so collisions between the hook and obstacles often occur.

[0003] In view of the above defects, existing technologies not only have manual monitoring, but also use sensor means for tower crane collision warning, but there are still problems of instability, difficulty in timely and accurate judgment and warning. For example: the detection distance of existing lidar sensors decreases in bad weather, and they cannot identify slender objects such as lifting ropes; a single vision or depth perception device cannot simultaneously obtain image semantic information and accurate three-dimensional dimensions.

[0004] In order to solve the above existing problems, people have been seeking an ideal technical solution. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned defects in the prior art, and provide a tower crane collision detection method based on multi-source perception, which effectively solves the collision warning problem in tower crane operation and improves the operation efficiency of tower cranes.

[0006] To achieve the above purpose, the present invention provides a tower crane collision detection method based on multi-source perception, including the following steps:

[0007] Step S1: Deploy a multi-source perception unit, the multi-source perception unit includes a visible light camera and a depth camera, the visible light camera collects image data of an object, and the depth camera collects depth information of the object;

[0008] Step S2: Contour detection technology, through the above-mentioned collected image data and depth information, process to form the contour of the object, and calculate the contour ratio of the hook and the suspended load;

[0009] Step S3: Calibrate the size of the object, establish a pixel equivalent based on the imaging size of the hook, and directly apply this proportional coefficient to calculate the true size of the unknown suspended load, so as to output the three-dimensional size of the suspended load;

[0010] Step S4: Construct the hybrid bounding box. During the operation of the tower crane, utilize the spatial pose information to construct a direction bounding box for the objects moving with the tower crane and an axis-aligned bounding box for the obstacles that remain stationary in the environment; for non-traditional square objects, construct a triangular prism bounding box.

[0011] Step S5: Output the collision situation. Deploy a parallel collision detection algorithm based on the separating axis theorem. The normal of each face of each bounding box provides a candidate separating axis. Through the analysis of the spatial topological relationship of the full-scene bounding box, set two levels of collision thresholds: when the penetration depth ≥ 0.2 m, trigger a warning signal; when the penetration depth ≥ 0.5 m, trigger an emergency stop signal.

[0012] Preferably, in the step S1, the visible light camera is installed on the luffing trolley, the depth camera is installed on the standard section of the tower body, the viewing angle of the visible light camera faces downward, the viewing angle of the depth camera is parallel to the ground, and the optical axis of the visible light camera is perpendicular to the optical axis of the depth camera.

[0013] Preferably, the visible light camera is shock-mounted on the luffing trolley, the depth camera is shock-mounted on the standard section of the tower body, and the visible light camera and the luffing trolley are connected by a first universal damping pan-tilt head, and the depth camera and the standard section of the tower body are connected by a second universal damping pan-tilt head.

[0014] Preferably, the resolution of the visible light camera for taking pictures is not less than 512×512; the depth camera adopts the time-of-flight ranging principle, with a ranging error of ±1.5 cm within a range of 10 m, and sets a depth threshold of 10%.

[0015] Preferably, the degrees of freedom of the first universal damping pan-tilt head and the second universal damping pan-tilt head are pitch ±30°, roll ±15°, and yaw ±45°.

[0016] Preferably, in the step S2, adopt the contour detection technology based on OpenCV; specifically include the following steps:

[0017] Step S21: Obtain the image data and depth information.

[0018] Step S22: Preprocess the image data; including grayscale processing and Gaussian blur. The grayscale processing uses cv2.cvtColor; the kernel size of the Gaussian blur operation is selected as 5×5, σ = 1.5.

[0019] Step S23: Perform Canny edge detection on the image data. The low threshold of the Canny edge detection is 50, and the high threshold is 100.

[0020] Step S24: Perform dilation operation, contour search, and contour screening on the image data. The dilation kernel for the dilation operation is selected as 3×3 and iterated once.

[0021] Step S25: Draw the contour of the object and calculate the contour ratio of the hook and the load.

[0022] Preferably, in step S4, the acquisition of the spatial pose information of the object to be collided includes acquiring the spatial poses of the hook, the tower crane slewing mechanism, the tower crane luffing mechanism, and the tower crane hoisting mechanism. The rotation angle of the tower crane slewing mechanism is acquired through an absolute encoder with an accuracy of ±0.1°. The displacement of the tower crane luffing mechanism is acquired through a laser rangefinder with an accuracy of ±1 cm within 100 m. The displacement of the tower crane hoisting mechanism is acquired through a heavy-duty encoder with an accuracy of ±2 cm within 200 m, and the pose update frequency is ≥20 Hz.

[0023] Preferably, in step S5, the warning signal is an audible and visual warning composed of a buzzer and LED flashing; the emergency stop measure is to cut off the hydraulic power of the tower crane.

[0024] Preferably, in step S4, the type of bounding box is selected according to the actual movement of the object to be established in the tower crane working environment. For the objects that move along with the tower crane during its operation, the load and the tower crane boom construct a directional bounding box. For the obstacles that remain stationary in the environment, the tower crane standard section constructs an axis-aligned bounding box. For non-traditional square objects, the cable constructs a triangular prism bounding box;

[0025] The calculation steps of the directional bounding box are as follows: First, calculate the local coordinate values of the eight vertices of the bounding box, and then convert them to the global coordinate system; it is expressed as (X, Y, Z) = T(R, d) * (x, y, z), where X, Y, Z are the global coordinates; x, y, z are the local coordinates; R = Rot(β), d = (Xc, Yc, Zc), β is the rotation angle of the bounding box, and c is the origin of the local coordinate system;

[0026] The coordinate conversion of the triangular prism bounding box is similar to the calculation steps of the directional bounding box, with the difference being the number of vertices;

[0027] The calculation method of the axis-aligned bounding box is: directly obtained through coordinate translation; it is expressed as (X, Y, Z) = (Xc, Yc, Zc) + (x, y, z).

[0028] Preferably, the triangular prism bounding box regards the bounding box of the cable part as a triangular prism, and the generation method of the triangular prism bounding box includes the following steps:

[0029] Step S91: defining the bottom triangle: taking the projection of the cable on the longitudinal section of the tower crane arm as the bottom triangle, the coordinates of the three vertices of the bottom triangle can be calculated based on the spatial coordinates of the tower crane's luffing mechanism, the length of the cable and the angle with the horizontal plane;

[0030] Step S92: Calculate the height direction: take the thickness of the cable as the height of the triangular prism, and the direction is perpendicular to the longitudinal section of the tower crane arm;

[0031] Step S93: generating top surface vertices: translating the vertices of the bottom surface triangle in the height direction to obtain corresponding vertices of the top surface triangle;

[0032] Step S94: Geometric modeling and coordinate transformation: connect the corresponding vertices of the bottom and top surfaces to generate the six rectangular sides of the triangular prism, forming a closed three-dimensional bounding box; obtain the rotation matrix and translation matrix through the tower crane rotation mechanism and the tower crane amplitude adjustment mechanism, and transform the local coordinate system of the triangular prism into the global coordinate system to adapt to the dynamic working scene of the tower crane.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The present invention combines a visible light camera with a depth camera to replace the traditional laser radar, effectively solving the problem of being unable to identify slender objects, such as lifting ropes, due to insufficient point cloud density. In addition, under severe weather conditions, the maximum detection distance of the laser radar will drop significantly, while the visible light camera passively receives light signals, ensuring stability and reliability in various environments.

[0035] 2. The design of the multi-source perception unit of the present invention is to form a multi-source perception unit through a heterogeneous combination of a visible light camera and a depth camera; this design not only retains the semantic information of the image, but also avoids the matching error problem caused by traditional stereo vision. At the same time, it is more accurate and reliable when calculating the three-dimensional size of an object.

[0036] 3. Application of the hook size benchmark calibration method of the present invention: A hook size benchmark calibration method is proposed to solve the size ambiguity problem; this method establishes pixel equivalent through the known standard hook imaging size, and is directly applied to the calculation of the true size of the unknown load, thereby outputting the three-dimensional size of the load; this not only simplifies the complex calibration process of traditional stereo vision, but also improves the measurement efficiency and accuracy.

[0037] 4. The present invention constructs a hybrid bounding box and adopts a hybrid modeling method that combines axis-aligned bounding box (AABB), oriented bounding box (OBB) and triangular prism bounding box, which improves the model generation efficiency and reduces unnecessary computing resource consumption; sets two-level collision thresholds to improve safety and reduce false alarm rates; and can adjust the thresholds according to different working conditions to adapt to changing needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 Schematic diagram of the main structure and hardware deployment of the tower crane related to the present invention;

[0040] Figure 2 Schematic diagram of the mechanism of the tower crane related to the present invention;

[0041] Figure 3 Installation method diagram of the depth camera provided by the present invention;

[0042] Figure 4 Schematic diagram of the calculation of the oriented bounding box (OBB) related to the present invention;

[0043] Figure 5 Schematic diagram of the calculation of the axis-aligned bounding box (AABB) related to the present invention;

[0044] Figure 6 Schematic diagram of the calculation of the triangular prism bounding box related to the present invention.

[0045] In the figure, it includes:

[0046] 1. Concrete foundation; 2. Standard tower section; 3. Tower crane arm; 4. Depth camera; 5. Visible light camera; 6. Luffing trolley; 7. Hook; 8. Load; 9. Second universal damping pan-tilt; 10. First universal damping pan-tilt; 11. Tower crane hoisting mechanism; 12. Tower crane slewing mechanism; 13. Tower crane luffing mechanism; 15. Depth camera optical axis; 16. Visible light camera optical axis; 17. Laser rangefinder; 18. Absolute encoder; 19. Heavy-duty encoder; 20. Cable. Detailed implementation manners

[0047] The following will clearly and completely describe the technical solutions in the present embodiment of the present invention in conjunction with the drawings in the present embodiment of the present invention. Obviously, the described present embodiment is an embodiment of the present invention, rather than all embodiments of the present invention. Based on the present embodiment of the present invention, all other present embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0048] Embodiment 1

[0049] Please refer to Figures 1 to 6,The present invention provides a tower crane collision detection method based on multi-source perception.

[0050] The following steps are involved:

[0051] Step S1: deploying a multi-source sensing unit, wherein the multi-source sensing unit includes a visible light camera 5 and a depth camera 4, wherein the visible light camera 5 collects image data of an object, and the depth camera 4 collects depth information of the object;

[0052] like Figure 1 As shown, in step S1, the visible light camera 5 is installed on the luffing trolley 6, and the depth camera 4 is installed on the standard section 2 of the tower body. The viewing angle of the visible light camera 5 is downward, and the viewing angle of the depth camera 4 is parallel to the ground. The visible light camera optical axis 16 of the visible light camera 5 and the depth camera optical axis 15 of the depth camera 4 are perpendicular to each other, which can ensure the integrity of the spatial data.

[0053] like Figure 1 and Figure 3 As shown, the visible light camera 5 is shock-absorbingly mounted on the luffing trolley 6, and the depth camera 4 is shock-absorbingly mounted on the tower standard section 2; the shock-absorbing installation can solve the problem of high-frequency micro-vibration and sudden shaking in tower crane operation affecting the camera.

[0054] Furthermore, in this embodiment, in order to maintain stable observation, reduce the impact of vibration caused by the tower crane during operation, and increase the life of the camera, the visible light camera 5 and the boom trolley 6 are connected via a first universal damping pan-tilt head 10, and the depth camera 4 and the tower standard section 2 are connected via a second universal damping pan-tilt head 9.

[0055] Furthermore, in this embodiment, the degrees of freedom of the first gimbal damping platform 10 and the second gimbal damping platform 9 are pitch ±30°, roll ±15° and yaw ±45°.

[0056] In order to ensure the accuracy of the calculated sizes of the hook 7 and the load 8, in the multi-source sensing unit: the resolution of the visible light camera 5 is not less than 512×512; at the same time, the depth camera 4 adopts the time of flight (ToF) ranging principle, with a ranging error of ±1.5cm within a range of 10m, and a depth threshold of 10% is set to save unnecessary computing resources.

[0057] Step S2: contour detection technology, through the above-collected image data and depth information, the contour of the object is processed and the contour ratio of the hook 7 and the suspended object 8 is calculated;

[0058] In step S2 described herein, the contour detection technology based on OpenCV is adopted; the contours of the lifting hook 7 and the suspended load 8 are accurately processed, and the contour ratios of the lifting hook 7 and the suspended load 8 can be calculated; data preparation is made for subsequent accurate calculation.

[0059] Specifically, it includes the following steps:

[0060] Step S21: Obtain the object image data collected by the visible light camera 5 and the object depth information collected by the depth camera 4;

[0061] Step S22: Preprocess the image data; including grayscale processing and Gaussian blur. The grayscale processing uses cv2.cvtColor; the kernel size of the Gaussian blur operation is selected as 5×5, and σ = 1.5;

[0062] Step S23: Perform Canny edge detection on the image data. The low threshold of the Canny edge detection is 50 and the high threshold is 100;

[0063] Step S24: Perform dilation operation, contour search and contour screening on the image data. The dilation kernel of the dilation operation is selected as 3×3 and iterated 1 time;

[0064] Step S25: Draw the contour of the object and calculate the contour ratios of the lifting hook 7 and the suspended load 8.

[0065] Step S3: Calibrate the size of the object, establish the pixel equivalent based on the imaging size of the lifting hook 7, and directly apply this proportional coefficient to calculate the real size of the unknown suspended load 8, so as to output the three-dimensional size of the suspended load 8;

[0066] In this embodiment, the size calculation part of the suspended load 8 is implemented by using the contour detection technology based on OpenCV and the calibration method of the size reference of the lifting hook 7, which can simplify the calibration process and ensure the reliability of the measurement.

[0067] In this embodiment, the size of the lifting hook 7 can adopt a standard fixed size, which can be pre-stored inside the software; of course, different suspended loads 8 may require different specifications of the lifting hook 7. All the standard series sizes of the lifting hook 7 are pre-stored inside the software, and only the specification of the lifting hook 7 in use needs to be selected, thus solving the problem of inaccurate size of the suspended load 8.

[0068] In this embodiment, the calibration method of the size reference of the lifting hook 7 establishes the pixel equivalent based on the imaging size of the lifting hook 7, and directly applies this proportional coefficient to calculate the real size of the unknown suspended load 8. The complex calibration process of traditional stereo vision can be avoided.

[0069] Step S4: Construct a hybrid bounding box. During the operation of the tower crane, utilize the spatial pose information to construct a directional bounding box for the object moving with the tower crane and an axis-aligned bounding box for the obstacles that remain stationary in the environment.

[0070] In step S4, the acquisition of the spatial pose information of the object to be collided includes obtaining the spatial poses of the hook 7, the tower crane slewing mechanism 12, the tower crane luffing mechanism 13, and the tower crane hoisting mechanism 11. The pose of the hook 7 is obtained by combining the visible light camera 5 and the depth camera 4. The rotation angle of the tower crane slewing mechanism 12 is obtained by the absolute encoder 18 with an accuracy of ±0.1°. The displacement of the tower crane luffing mechanism 13 is obtained by the laser rangefinder 17 with an accuracy of ±1 cm within 100 m. The displacement of the tower crane hoisting mechanism 11 is obtained by the heavy-duty encoder 19 with an accuracy of ±2 cm within 200 m, and the pose update frequency is ≥20 Hz.

[0071] In this embodiment, in step S4, the method for constructing the hybrid bounding box selects the type of bounding box according to the actual movement of the object to be established in the tower crane working environment to reduce computing resources. For the objects moving with the tower crane during its operation, the suspended load 8 and the tower crane boom 3 construct directional bounding boxes. For the obstacles that remain stationary in the environment, the tower standard section 2 constructs axis-aligned bounding boxes. For non-traditional square objects, the cable 20 constructs a triangular prism bounding box.

[0072] The calculation steps of the directional bounding box are as follows: First, calculate the local coordinate values of the eight vertices of the bounding box, and then convert them to the global coordinate system; expressed as (X, Y, Z) = T(R, d)*(x, y, z), where X, Y, Z are the global coordinates; x, y, z are the local coordinates; R = Rot(β), d = (Xc, Yc, Zc), β is the rotation angle of the bounding box, and c is the origin of the local coordinate system.

[0073] The coordinate conversion of the triangular prism bounding box is similar to the calculation steps of the directional bounding box, with the difference being the number of vertices.

[0074] The calculation method of the axis-aligned bounding box is: directly obtained by coordinate translation. Expressed as (X, Y, Z) = (Xc, Yc, Zc)+(x, y, z).

[0075] The triangular prism bounding box regards the bounding box of the cable 20 part as a triangular prism, and the generation method of the triangular prism bounding box includes the following steps:

[0076] Step S91: Definition of the bottom triangle: Use the projection of the cable 20 on the longitudinal section of the tower crane boom as the bottom triangle. Based on the spatial coordinates of the tower crane luffing mechanism 13, the length of the cable 20, and the angle with the horizontal plane, the coordinates of the three vertices of the bottom triangle can be calculated.

[0077] Step S92: Height direction calculation: Take the thickness of the cable 20 as the height of the triangular prism, and the direction is perpendicular to the longitudinal section of the tower crane arm 3.

[0078] Step S93: Top surface vertex generation: Translate the vertices of the bottom triangle along the height direction to obtain the corresponding vertices of the top surface triangle.

[0079] Step S94: Geometric modeling and coordinate transformation: Connect the corresponding vertices of the bottom surface and the top surface to generate six rectangular sides of the triangular prism to form a closed three-dimensional bounding box; obtain the rotation matrix and translation matrix through the tower crane slewing mechanism 12 and the tower crane luffing mechanism 13, and convert the local coordinate system of the triangular prism to the global coordinate system to adapt to the dynamic working scenario of the tower crane.

[0080] Step S5: Output the collision situation. A parallel collision detection algorithm based on the separating axis theorem is deployed. The normal of each face of each bounding box provides a candidate separating axis. By analyzing the spatial topological relationship of the full-scene bounding box, two-level collision thresholds are set: When the penetration depth ≥ 0.2m, a warning signal is triggered; when the penetration depth ≥ 0.5m, an emergency stop signal is triggered, which can improve safety and reduce the false alarm rate, and at the same time can dynamically adapt to different working conditions by adjusting the threshold.

[0081] In this embodiment, the triggering response measures for the two-level collision threshold are: The warning signal is an audible and visual warning, composed of a buzzer and LED flashing; the emergency stop measure is to cut off the hydraulic power of the tower crane.

[0082] For the collision detection between traditional bounding boxes (such as AABB and AABB, AABB and OBB, OBB and OBB), there are a total of 15 fixed candidate separating axes. In contrast, the total number of candidate separating axes for the collision detection between a polygonal bounding box and a traditional bounding box is not fixed. The specific calculation formula is: Total number of candidate separating axes = F + 3*E + 3, where F: the number of faces of the bounding box, which is a triangular prism in this embodiment, 3: the three main axis directions of the OBB / AABB (corresponding to the three orthogonal face normals), 3*E: the cross-product combination of the direction of each edge of the bounding box and the three main axis directions of the OBB (E is the number of edges of the bounding box).

[0083] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A tower crane collision detection method based on multi-source perception, characterized in that: It includes the following steps: Step S1: Deploy a multi-source sensing unit, which includes a visible light camera (5) and a depth camera (4). The visible light camera (5) acquires image data of an object, and the depth camera (4) acquires depth information of the object; Step S2: Contour detection technology. Through the acquired image data and depth information, the contour of the object is processed and formed, and the contour ratio of the hook (7) and the suspended load (8) is calculated; Step S3: Calibrate the size of the object. Establish a pixel equivalent based on the imaging size of the hook (7), and directly apply this proportional coefficient to calculate the real size of the unknown suspended load (8), so as to output the three-dimensional size of the suspended load (8); Step S4: Construct a hybrid bounding box. During the operation of the tower crane, use the spatial pose information to construct a direction bounding box for the object moving with the tower crane, and construct an axis-aligned bounding box for the obstacles that remain stationary in the environment; for non-traditional square objects, construct a triangular prism bounding box; Step S5: Output the collision situation. Deploy a parallel collision detection algorithm based on the separating axis theorem. Each face normal of each bounding box provides a candidate separating axis. Through the analysis of the spatial topological relationship of the full-scene bounding box, set two-level collision thresholds: when the penetration depth ≥ 0.2m, trigger a warning signal; when the penetration depth ≥ 0.5m, trigger an emergency stop signal.

2. The tower crane collision detection method based on multi-source perception according to claim 1, characterized in that: In the said step S1, the visible light camera (5) is installed on the luffing trolley (6), the depth camera (4) is installed on the tower standard section (2), the viewing angle of the visible light camera (5) faces downward, the viewing angle of the depth camera (4) is parallel to the ground, and the optical axis (16) of the visible light camera of the visible light camera (5) is perpendicular to the optical axis (15) of the depth camera of the depth camera (4).

3. The method for detecting tower crane collisions based on multi-source perception according to claim 2, characterized in that: The visible light camera (5) is shock-absorbingly installed on the luffing trolley (6), the depth camera (4) is shock-absorbingly installed on the tower standard section (2), and the visible light camera (5) and the luffing trolley (6) are connected through a first universal damping pan-tilt head (10), and the depth camera (4) and the tower standard section (2) are connected through a second universal damping pan-tilt head (9).

4. The tower crane collision detection method based on multi-source perception according to claim 3, wherein: The resolution of the photo taken by the visible light camera (5) is not less than 512×512; the depth camera (4) adopts the time-of-flight ranging principle, with a ranging error of ±1.5cm within a range of 10m, and a depth threshold of 10% is set.

5. The tower crane collision detection method based on multi-source perception according to claim 3, characterized in that: The degrees of freedom of the first universal damping pan-tilt head (10) and the second universal damping pan-tilt head (9) are pitch ±30°, roll ±15° and yaw ±45°.

6. The tower crane collision detection method based on multi-source perception according to claim 1, characterized in that: In the said step S2, an OpenCV-based contour detection technology is adopted; specifically, it includes the following steps: Step S21: Obtain image data and depth information; Step S22: Preprocess the image data; including grayscale processing and Gaussian blur. The grayscale processing uses cv2.cvtColor; the kernel size of the Gaussian blur operation is selected as 5×5, σ = 1.5; Step S23: Perform Canny edge detection on the image data. The low threshold of the Canny edge detection is 50, and the high threshold is 100; Step S24: Perform dilation operation, contour finding, and contour screening on the image data. The dilation kernel for the dilation operation is a 3×3 kernel, and the iteration is performed once. Step S25: Draw the contour of the object and calculate the contour ratio of the hook (7) and the suspended load (8).

7. A tower crane collision detection method based on multi-source perception according to claim 1, characterized in that: In step S4, the acquisition of the spatial pose information of the object to be collided includes obtaining the spatial poses of the hook (7), the tower crane slewing mechanism (12), the tower crane luffing mechanism (13), and the tower crane hoisting mechanism (11). The rotation angle of the tower crane slewing mechanism (12) is obtained through an absolute encoder (18) with an accuracy of ±0.1°; the displacement of the tower crane luffing mechanism (13) is obtained through a laser rangefinder (17) with an accuracy of ±1 cm within 100 m; the displacement of the tower crane hoisting mechanism (11) is obtained through a heavy-duty encoder (19) with an accuracy of ±2 cm within 200 m, and the pose update frequency ≥ 20 Hz.

8. A tower crane collision detection method based on multi-source perception according to claim 1, characterized in that: In step S5, the warning signal is an audible and visual warning composed of a buzzer and LED flashing; the emergency stop measure is to cut off the hydraulic power of the tower crane.

9. A tower crane collision detection method based on multi-source perception according to claim 7, characterized in that: In step S4, the method for constructing the hybrid bounding box selects the type of bounding box according to the actual motion of the object to be established in the working environment of the tower crane. For the objects that move along with the tower crane during its operation, the suspended load (8) and the tower crane boom (3) construct a directional bounding box; for the obstacles that remain stationary in the environment, the tower standard section (2) constructs an axis-aligned bounding box; for non-traditional square objects, the cable (20) constructs a triangular prism bounding box. The calculation steps of the directional bounding box are as follows: First, calculate the local coordinate values of the eight vertices of the bounding box, and then convert them to the global coordinate system; it is expressed as (X, Y, Z) = T(R, d)*(x, y, z), where X, Y, Z are the global coordinates; x, y, z are the local coordinates; R = Rot(β), d = (Xc, Yc, Zc), β is the rotation angle of the bounding box, and c is the origin of the local coordinate system. The coordinate conversion of the triangular prism bounding box is similar to the calculation steps of the directional bounding box, with the difference being the number of vertices. The calculation method of the axis-aligned bounding box is: directly obtained through coordinate translation; it is expressed as (X, Y, Z) = (Xc, Yc, Zc)+(x, y, z).

10. A tower crane collision detection method based on multi-source perception according to claim 9, characterized in that: The triangular prism bounding box regards the bounding box of the cable (20) part as a triangular prism. The generation method of the triangular prism bounding box includes the following steps: Step S91: Definition of the bottom triangle: Use the projection of the cable (20) on the longitudinal section of the tower crane boom (3) as the bottom triangle. Based on the spatial coordinates of the tower crane luffing mechanism (13), the length of the cable (20), and the angle with the horizontal plane, the coordinates of the three vertices of the bottom triangle can be calculated. Step S92: Calculation in the height direction: Use the thickness of the cable (20) as the height of the triangular prism, and the direction is perpendicular to the longitudinal section of the tower crane boom (3). Step S93: Generation of the top vertices: Translate the vertices of the bottom triangle along the height direction to obtain the corresponding vertices of the top triangle. Step S94: Geometric modeling and coordinate transformation: Connect the corresponding vertices of the bottom surface and the top surface to generate six rectangular side surfaces of the triangular prism, forming a closed three-dimensional bounding box; Obtain the rotation matrix and translation matrix through the tower crane slewing mechanism (12) and the tower crane luffing mechanism (13), and convert the local coordinate system of the triangular prism to the global coordinate system to adapt to the dynamic working scenario of the tower crane.

Images