A spraying track planning method of an intelligent adaptive helmet

Abstract

The application relates to the technical field, and particularly discloses a spraying track planning method of an intelligent adaptive helmet, which comprises the following steps: discretizing a spraying track S0 of a curved surface model M0 into a plurality of track points, and obtaining spraying data corresponding to the track points; calculating a rigid body transformation T0 of aligning the curved surface model M0 with a curved surface model M1, obtaining a preliminary coordinate, a preliminary track S1 and a mapping point based on the rigid body transformation T0; intercepting a spherical region of the curved surface model M1 based on the preliminary coordinate, and obtaining rigid body transformation parameters of the spherical region; training a nozzle pose correction model based on the mapping point and the corresponding rigid body transformation parameters; correcting the preliminary track S1 to obtain a corrected track S2 based on the nozzle pose correction model, performing a spraying distance compensation on the corrected track S2 to obtain a corrected track S3, and performing spraying on the helmet according to the corrected track S3. Through space mapping and local pose correction, the spraying track is synchronously fitted with the position and the posture of the curved surface, and the deviation caused by the positioning error is reduced.

Description

Technical Field

[0001] This invention relates to the field of spraying trajectory planning technology, and specifically to a spraying trajectory planning method for an intelligent adaptive helmet. Background Technology

[0002] Helmet painting is a surface treatment method that combines practicality and aesthetics. Through the combination of colors and patterns, helmets can showcase individuality and style while providing protection. The painted surface is smooth and delicate, with uniform and long-lasting color, which not only enhances visual recognition but also increases weather resistance and abrasion resistance to a certain extent.

[0003] In the production of helmets of the same model and batch, even if the product shape and design parameters are completely identical, tooling positioning deviations can still significantly affect the spraying trajectory. Positioning deviations may originate from various factors such as slight looseness in fixture assembly, clearance between the positioning pin and the workpiece, and differences in clamping force. These deviations are often in the millimeter or even sub-millimeter range, but for a process like spraying that is highly sensitive to spray distance, spray width overlap, and spray angle, they are enough to cause quality problems such as uneven coating thickness, localized missed spraying, or over-spraying. Summary of the Invention

[0004] The purpose of this invention is to provide a spraying trajectory planning method for an intelligent adaptive helmet, thereby solving the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for planning the spraying trajectory of an intelligent adaptive helmet includes the following steps:

[0007] A surface model M0 is constructed based on multi-view color images of a normal helmet and corresponding depth data. The spraying trajectory S0 of the surface model M0 is discretized into several trajectory points, and the spraying data corresponding to the trajectory points is obtained. The spraying data includes the three-dimensional coordinates and attitude vector of the nozzle.

[0008] Based on the multi-view color images of the current helmet and the corresponding depth data, construct a surface model M1. Calculate the rigid body transformation T0 to align the surface model M0 with the surface model M1 based on the ICP algorithm. Perform a rigid body transformation T0 on the three-dimensional coordinates of the nozzle to obtain the initial coordinates. Connect the initial coordinates to obtain the initial trajectory S1. Obtain the point in the surface model M1 that is closest to the initial coordinates and denote it as the mapping point.

[0009] Based on the initial coordinates, a spherical region is intercepted on the surface model M1, and the rigid body transformation parameters of the spherical region are obtained. The rigid body transformation parameters include the translation and rotation of the nozzle position. The nozzle pose correction model is trained based on the mapping points and the corresponding rigid body transformation parameters.

[0010] The initial trajectory S1 is corrected based on the nozzle pose correction model to obtain the corrected trajectory S2. The spray distance is compensated for the corrected trajectory S2 to obtain the corrected trajectory S3. The helmet is then painted according to the corrected trajectory S3.

[0011] As a further aspect of the present invention: the process of obtaining trajectory points includes:

[0012] Obtain color images and corresponding depth data of a normal helmet at m preset viewing angles, and reconstruct the surface model M0 based on TSDF;

[0013] Obtain the spraying trajectory S0 of the curved surface model MO, and discretize the spraying trajectory S0 into several trajectory points. The lengths of two adjacent trajectory points along the spraying trajectory S0 are the same and are preset values.

[0014] As a further aspect of the present invention: the process of obtaining the rigid body transformation parameters of the spherical region includes:

[0015] Starting from the beginning of the initial trajectory S1, a set number of adjacent initial coordinates on the initial trajectory S1 are grouped together. For a single group, a spherical region with a set radius is intercepted in the surface model M1 with the initial coordinates in the group as the center.

[0016] The spherical region is projected onto the multi-view color image of the current helmet to obtain a local image. Feature points of the local image are extracted based on the SIFT algorithm, and a corresponding set of feature points is established between different local images.

[0017] The corresponding point set is input into the random sampling consistency algorithm to eliminate incorrect pairs that do not meet geometric consistency, and geometrically consistent feature point pairs are obtained.

[0018] The position of the feature point pair in three-dimensional space and its two-dimensional projection position in the corresponding multi-view color image are input into the perspective n-point algorithm to calculate the rigid body transformation parameters of the corresponding spherical region. The rigid body transformation parameters include the translation and rotation of the nozzle position.

[0019] As a further aspect of the present invention, the process of obtaining the corrected trajectory S2 includes:

[0020] Using the coordinates A1 of the mapping point on the surface model M1 as input variables and the translation and rotation corresponding to coordinates A1 as output variables, a Gaussian process regression algorithm is used to generate a nozzle pose correction model that continuously changes on the surface model M1.

[0021] Input the coordinates of the mapping point into the nozzle pose correction model to obtain the corresponding translation and rotation amounts; apply the translation amount to the initial coordinates corresponding to the mapping point to obtain the corrected position coordinates.

[0022] The attitude vector of the nozzle corresponding to the initial coordinates is adjusted according to the rotation amount to obtain the corrected attitude vector;

[0023] Pair the corrected position coordinates with the corrected attitude vector to form corrected trajectory points, and connect the corrected trajectory points to obtain the corrected trajectory S2.

[0024] As a further aspect of the present invention, the process of obtaining the corrected trajectory S3 includes:

[0025] Using the position coordinates and attitude vector of each corrected trajectory point as input, the shortest distance from the nozzle position to the surface of the surface model M1 is calculated along the direction of the attitude vector. The difference between the shortest distance and the preset spraying target distance is used to obtain the distance deviation value.

[0026] Along the direction of the attitude vector, the position coordinates of the corrected trajectory points are translated axially according to the distance deviation value to obtain the trajectory points after spray distance compensation. The corrected trajectory S3 is generated by connecting the trajectory points after spray distance compensation.

[0027] As a further aspect of the present invention: the helmet painting according to the corrected trajectory S3 includes:

[0028] Based on the spatial distance and attitude change between the adjacent position coordinates and attitude vector pairing data on the corrected trajectory S3, a trajectory curve with continuous position and attitude is generated using spline interpolation. The trajectory curve is then time-annotated according to the speed model of the spraying process to obtain a nozzle pose sequence arranged in time order.

[0029] The spraying task is performed according to the nozzle pose sequence.

[0030] As a further aspect of the present invention: the nozzle pose correction model is trained using a backpropagation algorithm.

[0031] The beneficial effects of this invention compared to the prior art are as follows:

[0032] 1) This invention establishes a spatial mapping relationship between the standard spraying trajectory and the current spraying trajectory, and combines it with the pose correction of local areas, so that the nozzle can conform to the actual curved surface position and posture of the current helmet during the spraying process, realize the synchronous adjustment of the spraying trajectory in spatial position and spraying direction, reduce the spraying deviation caused by positioning error, and improve the uniformity and coverage integrity of the coating.

[0033] 2) This invention compensates for the spraying distance by correcting the spraying trajectory, so that the distance between the nozzle and the helmet surface is kept within the range that meets the spraying requirements, reducing the difference in paint deposition caused by the change in spraying distance, thereby reducing the risk of local coatings being too thick or too thin, making the coating surface more stable in thickness distribution, and the surface texture and gloss more consistent.

[0034] 3) By performing continuous processing and time labeling on the trajectory after spray distance compensation, this invention makes the movement of the nozzle on the spraying path smoother and more stable, reduces sudden changes in speed and attitude between trajectory points, improves the stability and controllability of the spraying process, ensures the consistency of the spraying process in terms of speed, direction and coverage, and improves the repeatability of spraying under mass production conditions. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a flowchart illustrating the spraying trajectory planning method for an intelligent adaptive helmet according to the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figure 1 As shown, this invention provides a method for planning the spraying trajectory of an intelligent adaptive helmet, comprising the following steps:

[0039] Step 1: Construct a curved surface model M0 based on the multi-view color image of a normal helmet and the corresponding depth data. Discretize the spraying trajectory S0 of the curved surface model M0 into several trajectory points and obtain the spraying data corresponding to the trajectory points. The spraying data includes the three-dimensional coordinates and attitude vector of the nozzle.

[0040] In a preferred embodiment of the present invention, the process of obtaining trajectory points includes:

[0041] A normal helmet, selected manually and in an ideal state without deformation or assembly deviation, is placed on a workstation for data acquisition. Multiple acquisition devices installed at different locations are used to photograph the helmet. Each acquisition device can simultaneously acquire color images and corresponding depth data. The color images record the color distribution information on the helmet surface, and the depth data records the spatial distance information from the acquisition device to each point on the helmet surface.

[0042] After acquiring color images and depth data from multiple different viewing angles, all depth data are aligned according to the spatial position at the time of acquisition. By mapping each depth pixel to its corresponding spatial position as a three-dimensional point and calculating the relative positional relationship of these points in space, combined with the principle of three-dimensional signed distance field reconstruction, a continuous surface representation of the helmet is generated. This surface model describes the shape features and geometric contours of the helmet's outer surface in space.

[0043] Plan the spraying trajectory S0 on the curved surface model. The trajectory is a continuous spatial path formed on the curved surface according to the spraying coverage requirements. The points passed by the path cover the target spraying area of ​​the helmet. The specific planning method is not limited.

[0044] After obtaining the trajectory S0, it is discretized at equal intervals according to the path length. The surface path length between each two adjacent discrete points remains consistent. The set of points obtained by this discretization is the trajectory point. Each trajectory point represents a fixed position on the trajectory. In order to obtain the spraying data, it is necessary to determine the position of each trajectory point in the coordinate system of the surface model, and determine the three-dimensional coordinates of the nozzle at that point and the nozzle attitude vector. The three-dimensional coordinates of the nozzle are the position of the trajectory point in space, and the attitude vector describes the orientation of the nozzle at that position, which is usually determined by the direction of the nozzle axis.

[0045] Step 2: Construct a surface model M1 based on the current multi-view color image of the helmet and the corresponding depth data. Calculate the rigid body transformation T0 to align the surface model M0 with the surface model M1 using the ICP algorithm. Perform the rigid body transformation T0 on the three-dimensional coordinates of the nozzle to obtain the initial coordinates. Connect the initial coordinates to obtain the initial trajectory S1. Obtain the point in the surface model M1 that is closest to the initial coordinates and record it as the mapping point.

[0046] In one specific embodiment, in this step, the current helmet to be painted is placed in the same working position as when collecting normal helmet data, and multiple acquisition devices with the same layout are used to acquire color images and corresponding depth data of the helmet from different directions. M1 is generated by analogy with the acquisition method of MO, which will not be elaborated here.

[0047] The standard surface model M0 and the current surface model M1 are input into the iterative nearest point algorithm. The algorithm first finds the closest point pairs in the two sets of surface data, and then calculates a rigid body transformation in three-dimensional space based on these point pairs. The rigid body transformation includes two parts: rotation and translation. The rotation part is used to adjust the orientation of the two to be consistent, and the translation part is used to adjust the spatial position of the two to be consistent. By continuously iterating the matching of the nearest point and the calculation of the rigid body transformation, the overall shape of M0 and M1 is made to overlap as much as possible, and finally the rigid body transformation T0 is obtained.

[0048] After obtaining T0, the three-dimensional coordinates of the nozzles corresponding to each trajectory point in the standard spraying trajectory S0 are substituted into T0 for transformation. The principle of transformation is to rotate the spatial position of each point according to the rotation matrix in T0, and then translate it according to the displacement vector in T0, so as to map these nozzle coordinates from the coordinate system of M0 to the coordinate system of M1. The transformed coordinates are called the preliminary coordinates.

[0049] Connect all the initial coordinates sequentially according to the spraying order of the original standard trajectory S0 to obtain the initial trajectory S1 in the current helmet coordinate system. In order to determine the accurate corresponding position of each position in the initial trajectory S1 on the current surface model M1, it is necessary to find the surface point closest to the initial coordinate in the surface data of M1. This process is accomplished by calculating the Euclidean distance between the initial coordinate and each point on the surface. The point with the smallest distance is the mapping point. In this way, each initial coordinate can find a corresponding physical position on the surface of M1, providing a basis for subsequent local analysis and trajectory correction.

[0050] Step 3: Based on the initial coordinates, a spherical region is extracted from the surface model M1, and the rigid body transformation parameters of the spherical region are obtained. The rigid body transformation parameters include the translation and rotation of the nozzle position; the nozzle pose correction model is trained based on the mapping points and the corresponding rigid body transformation parameters.

[0051] In another preferred embodiment of the present invention, the process of obtaining the rigid body transformation parameters of the spherical region includes:

[0052] Starting from the initial trajectory S1, select several adjacent initial coordinates along the spraying sequence as a group. In the surface model M1, use each initial coordinate in the group as the center to cut out a spherical region with a preset radius. The spherical region is used to aggregate surface points and texture information near the coordinate. The sphere is used because this shape is not sensitive to curvature changes in three-dimensional space and can stably cover the local geometry around the same position.

[0053] The three-dimensional points in the spherical region are mapped onto the multi-view color image of the current helmet along the line of sight of each viewpoint to obtain the local image corresponding to the spherical region. The SIFT algorithm is applied to these local images to detect feature points with obvious local gray-level changes, and feature descriptors describing the texture pattern around each feature point are calculated. These descriptors are then paired between local images from different viewpoints. The pairing principle is to find a pair of pixel positions with the most similar texture patterns under different viewpoints, thereby forming a pair of feature points representing the same physical location.

[0054] Considering the possibility of mismatches in multi-view matching, the feature point pair set is further processed by the random sampling consensus algorithm. This algorithm assumes a geometric relationship by repeatedly sampling a small number of paired points and checks whether it can be satisfied by most paired points. Pairs that cannot be satisfied are judged as errors and removed, thus obtaining geometrically consistent feature point pairs.

[0055] For each pair of feature points that are retained, the corresponding 3D point in the spherical region is found by looking up the pixel position of the feature point in the local image. The reason why the 3D point and the pixel position are in one-to-one correspondence is that the known imaging geometry is used in this local area. The 3D point is projected onto the pixel of the image plane. When looking up the feature point, the 3D point with the most consistent correspondence is selected in the projection neighborhood.

[0056] The projection process here involves mapping the set of three-dimensional points obtained in the surface model M1 with the initial coordinates as the center of the sphere to the pixel positions on the acquired multi-view color image one by one:

[0057] For any given viewpoint, read the spatial position and orientation recorded during acquisition, and use this pose to transform each 3D point within the spherical region from the common coordinate system used by the surface model M1 to the camera's own coordinate system. The principle of this step is to rewrite the "object coordinates" as "coordinates under the camera's viewpoint", which makes it easier to locate the image plane according to the imaging relationship of that viewpoint.

[0058] Applying the pinhole imaging principle to the transformed 3D point, the 3D coordinates are mapped to the position of 2D pixels on the image according to the camera's imaging scale and imaging center, thus obtaining the landing point of the 3D point in the color image at that viewpoint. To avoid including occluded points in the local image, a visibility check is also required: for each projected pixel, compare the distance recorded by the pixel in the corresponding depth data at that viewpoint with the distance of the 3D point from the camera. If the depth data is closer, it means that the 3D point is occluded by the foreground and is removed, leaving only visible points with the same depth.

[0059] For the retained pixel positions, the minimum bounding boundary of the pixel coordinates is calculated. Within this boundary, a continuous image region is cropped from the original color image, which is the local image of the spherical region from this viewpoint. At the same time, the one-to-one correspondence between pixels and 3D points is stored internally, which facilitates the subsequent mutual referencing of image features and 3D positions. The above steps are repeated for all viewpoints, and each viewpoint will obtain one or more local images. If the projection points in a certain viewpoint cannot form an effective boundary region or only scattered pixels remain, no local image is generated for that viewpoint. Finally, the set of these cropped local images, together with their correspondence with the 3D points of the spherical region, is used as the input for subsequent feature extraction, matching, and pose solving.

[0060] The obtained 3D points and their pixel positions in the multi-view color image are input into the PnP algorithm. PnP uses several 3D-2D correspondences to solve the rotation and displacement of the local region relative to the imaging coordinate system. The obtained rotation and displacement can be understood as the local attitude and position offset introduced by the clamping deviation in the spherical region, and then used as the rigid body transformation parameters of the spherical region. The rotation amount is used to characterize the angle that the nozzle attitude needs to be corrected, and the translation amount is used to characterize the displacement that the nozzle position needs to be corrected.

[0061] Following the above method, all groups are processed sequentially along the initial trajectory S1. Rigid body transformation parameters for each spherical region are collected, and these parameters are paired with the corresponding mapping points. The position of the mapping point in the surface model M1 is used as input, and the translation and rotation of the corresponding spherical region are used as output. Training samples covering the entire initial trajectory S1 are compiled. By fitting these samples, a nozzle pose correction model is obtained. This model provides a continuous mapping from any mapping point position to the corresponding position and attitude correction amount on the surface model M1, which is used for subsequent point-by-point correction of the initial trajectory S1.

[0062] It should be noted that in order to continuously predict position and orientation separately, the position coordinates of all mapped points and their corresponding translations are organized into the first set of training samples, and the position coordinates of the mapped points and their corresponding rotations are organized into the second set of training samples.

[0063] The training principle is to construct a continuously defined functional relationship on the surface model M1 based on these discrete samples, so that any input of the position of the mapping point on the surface model M1 can output the corresponding translation or rotation amount through the functional relationship.

[0064] Since position correction and translation correction are different physical quantities with different dimensions and laws of change, two independent prediction models are established: one to output the translation amount and the other to output the rotation amount. After training, the first model (the model that outputs the translation amount) can predict the translation correction value of the nozzle at any position on the surface model M1, and the second model (the model that outputs the rotation amount) can predict the attitude rotation correction value of the nozzle at the same position. The combination of the two can provide complete pose correction information for the subsequent trajectory correction steps.

[0065] It is understandable that the nozzle pose correction model is trained using the backpropagation algorithm. The specific training method is already quite mature in the current technology, so it will not be described in detail here.

[0066] In a preferred embodiment of this invention, the process of obtaining the corrected trajectory S2 includes:

[0067] The 3D position A1 of the mapping point in the surface model M1 is used as input to the trained nozzle pose correction model. The nozzle pose correction model consists of two parts: one is used to predict the translation correction amount of the position, and the other is used to predict the rotation correction amount of the pose. Both models are trained with the mapping point position as the input variable. Therefore, when A1 is input, the first model outputs the 3D translation amount that needs to be performed at that position, and the second model outputs the rotation amount that needs to be performed at that position.

[0068] After obtaining the translation amount, it is directly applied to the initial coordinates corresponding to the mapping point. Specifically, the three-dimensional position vector of the initial coordinates is added to the translation vector component by component. The principle of addition is that the translation amount itself describes the displacement correction value in the three spatial axis directions. Therefore, adding it to the initial coordinates will give the corrected position coordinates.

[0069] The nozzle attitude vector corresponding to the initial coordinates is adjusted according to the rotation amount. The three components of the rotation amount correspond to the rotation angle of the nozzle in the three spatial axis directions. During the adjustment, the attitude vector is transformed in the coordinate sequence according to the rotation order defined by the rotation amount, so that the attitude direction is rotated in the three-dimensional space to obtain the corrected attitude vector.

[0070] After completing the position and attitude correction, the corrected position coordinates and the corrected attitude vectors are paired and stored. This pairing data represents the complete spatial pose of the nozzle at the spraying position. According to the order of the preliminary coordinates in the preliminary trajectory S1, all the corrected position and attitude pairing data (corrected trajectory points) are connected sequentially to generate the corrected trajectory S2. This trajectory retains the path order of the preliminary trajectory in space, but each position and direction has been adjusted by the nozzle pose correction model to fit the actual situation of the current surface model M1.

[0071] Step 4: Based on the nozzle pose correction model, correct the initial trajectory S1 to obtain the corrected trajectory S2, perform spray distance compensation on the corrected trajectory S2 to obtain the corrected trajectory S3, and spray the helmet according to the corrected trajectory S3.

[0072] In another preferred embodiment of the present invention, the process of obtaining the corrected trajectory S3 includes:

[0073] Each point in the corrected trajectory S2 is extracted sequentially, and its three-dimensional position coordinates and corresponding attitude vector are read. These two coordinates are then used as inputs for calculating the jet distance deviation.

[0074] In the surface model M1, starting from the position of the correction trajectory point, a spatial ray is established along the direction of the attitude vector. This ray will intersect with the outer surface of the surface model M1. The straight-line distance from the intersection point to the starting point is the current spray distance of the nozzle. The principle of spray distance calculation here is that the attitude vector is the spatial representation of the nozzle spray direction. Using it as the ray direction can directly simulate the path of paint flow when the nozzle is spraying.

[0075] The calculated current spray distance is compared with the preset spray target distance. The difference between the two gives the distance deviation value. A positive value indicates that the nozzle is farther than the target distance, and a negative value indicates that the nozzle is closer than the target distance.

[0076] Along the direction of the attitude vector, the position coordinates of the correction trajectory point are adjusted according to the distance deviation value. If the deviation is positive, the position is translated in the opposite direction of the attitude vector to shorten the spray distance. If the deviation is negative, the position is translated in the positive direction of the attitude vector to increase the spray distance. The principle of translation is to add a displacement vector in space along the direction or opposite direction of the attitude vector, thereby changing the distance between the nozzle and the curved surface.

[0077] After this translation operation, each corrected trajectory point will generate a new position coordinate. This coordinate is paired with the original attitude vector to form a trajectory point after spray distance compensation. According to the order of the trajectory points in the corrected trajectory S2, all the trajectory points after spray distance compensation are connected in sequence to form the corrected trajectory S3. This trajectory not only retains the shape of S2 on the spatial path, but also makes the distance between the nozzle and the curved surface closer to the target spraying conditions through position adjustment.

[0078] It should be noted that the helmet painting process based on the corrected trajectory S3 includes:

[0079] The position coordinates of each nozzle distance after compensation and the corresponding attitude vector in the corrected trajectory S3 are read sequentially according to the trajectory order. The spatial distance between two adjacent paired data and the degree of difference in attitude vectors are calculated. The spatial distance reflects the change range of nozzle position and the attitude difference reflects the change range of nozzle injection direction.

[0080] Based on this data, spline interpolation is used to generate the position and attitude of intermediate points between adjacent paired data. The position coordinates of the intermediate points are obtained by smoothing the three-dimensional spatial coordinates, and the attitude vector is obtained by interpolating the three-dimensional direction vector. The principle of interpolation is to calculate the value at any position between two points based on the values ​​and position ratios of adjacent known points, so that the position and attitude changes form a continuous curve on the trajectory, rather than being directly connected by a series of discrete points. The trajectory curve generated in this way is smooth and continuous on the spatial path, and also maintains a gradual transition in attitude changes, avoiding abrupt changes during nozzle movement.

[0081] The trajectory curve is combined with the speed model of the spraying process, and specific time labels are assigned to the pairing data of each position and attitude on the curve. The time labeling process is to allocate the time that the nozzle should stay or pass through at different positions according to the spraying requirements and the amplitude of position and attitude changes on the trajectory, so as to ensure the uniformity and continuity of the spraying. The labeling process can be based on manual or other methods, and there are no restrictions here.

[0082] After time labeling is completed, the trajectory curve is transformed into a set of nozzle pose sequences arranged in chronological order. Each item in the sequence contains the three-dimensional position coordinates and injection attitude vector of the nozzle at a specific moment.

[0083] The nozzle pose sequence is sent to the spraying actuator. The spraying actuator will drive the nozzles to the designated positions and maintain the designated postures in the time sequence to perform spraying until the entire trajectory is completed.

[0084] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for planning the spraying trajectory of an intelligent adaptive helmet, characterized in that, Includes the following steps: A surface model M0 is constructed based on multi-view color images of a normal helmet and corresponding depth data. The spraying trajectory S0 of the surface model M0 is discretized into several trajectory points, and the spraying data corresponding to the trajectory points is obtained. The spraying data includes the three-dimensional coordinates and attitude vector of the nozzle. Based on the multi-view color images of the current helmet and the corresponding depth data, construct a surface model M1. Calculate the rigid body transformation T0 to align the surface model M0 with the surface model M1 based on the ICP algorithm. Perform a rigid body transformation T0 on the three-dimensional coordinates of the nozzle to obtain the initial coordinates. Connect the initial coordinates to obtain the initial trajectory S1. Obtain the point in the surface model M1 that is closest to the initial coordinates and denote it as the mapping point. Based on the initial coordinates, a spherical region is intercepted on the surface model M1, and the rigid body transformation parameters of the spherical region are obtained. The rigid body transformation parameters include the translation and rotation of the nozzle position. The nozzle pose correction model is trained based on the mapping points and the corresponding rigid body transformation parameters. The initial trajectory S1 is corrected based on the nozzle pose correction model to obtain the corrected trajectory S2. The spray distance is compensated for the corrected trajectory S2 to obtain the corrected trajectory S3. The helmet is then painted according to the corrected trajectory S3.

2. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 1, characterized in that, The process of obtaining trajectory points includes: Obtain color images and corresponding depth data of a normal helmet at m preset viewing angles, and reconstruct the surface model M0 based on TSDF; Obtain the spraying trajectory S0 of the curved surface model MO, and discretize the spraying trajectory S0 into several trajectory points. The lengths of two adjacent trajectory points along the spraying trajectory S0 are the same and are preset values.

3. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 2, characterized in that, The process of obtaining the rigid body transformation parameters of the spherical region includes: Starting from the beginning of the initial trajectory S1, a set number of adjacent initial coordinates on the initial trajectory S1 are grouped together. For a single group, a spherical region with a set radius is intercepted in the surface model M1 with the initial coordinates in the group as the center. The spherical region is projected onto the multi-view color image of the current helmet to obtain a local image. Feature points of the local image are extracted based on the SIFT algorithm, and a corresponding set of feature points is established between different local images. The corresponding point set is input into the random sampling consistency algorithm to eliminate incorrect pairs that do not meet geometric consistency, and geometrically consistent feature point pairs are obtained. The position of the feature point pair in three-dimensional space and its two-dimensional projection position in the corresponding multi-view color image are input into the perspective n-point algorithm to calculate the rigid body transformation parameters of the corresponding spherical region. The rigid body transformation parameters include the translation and rotation of the nozzle position.

4. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 3, characterized in that, The process of obtaining the corrected trajectory S2 includes: Using the coordinates A1 of the mapping point on the surface model M1 as input variables and the translation and rotation corresponding to coordinates A1 as output variables, a Gaussian process regression algorithm is used to generate a nozzle pose correction model that continuously changes on the surface model M1. Input the coordinates of the mapping point into the nozzle pose correction model to obtain the corresponding translation and rotation amounts; apply the translation amount to the initial coordinates corresponding to the mapping point to obtain the corrected position coordinates. The attitude vector of the nozzle corresponding to the initial coordinates is adjusted according to the rotation amount to obtain the corrected attitude vector; Pair the corrected position coordinates with the corrected attitude vector to form corrected trajectory points, and connect the corrected trajectory points to obtain the corrected trajectory S2.

5. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 4, characterized in that, The process of obtaining the corrected trajectory S3 includes: Using the position coordinates and attitude vector of each corrected trajectory point as input, the shortest distance from the nozzle position to the surface of the surface model M1 is calculated along the direction of the attitude vector. The difference between the shortest distance and the preset spraying target distance is used to obtain the distance deviation value. Along the direction of the attitude vector, the position coordinates of the corrected trajectory points are translated axially according to the distance deviation value to obtain the trajectory points after spray distance compensation. The corrected trajectory S3 is generated by connecting the trajectory points after spray distance compensation.

6. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 5, characterized in that, The helmet painting process based on the corrected trajectory S3 includes: Based on the spatial distance and attitude change between the adjacent position coordinates and attitude vector pairing data on the corrected trajectory S3, a trajectory curve with continuous position and attitude is generated using spline interpolation. The trajectory curve is then time-annotated according to the speed model of the spraying process to obtain a nozzle pose sequence arranged in time order. The spraying task is performed according to the nozzle pose sequence.

7. The spraying trajectory planning method for an intelligent adaptive helmet according to claim 6, characterized in that, The nozzle pose correction model is trained using the backpropagation algorithm.

Images