Adjustable mobile phone shell laser cutting table

Abstract

The invention discloses an adjustable mobile phone shell laser cutting table, which relates to the technical field of self-adaptive control and comprises a base platform, an adsorption fixing unit, a visual acquisition unit, a laser execution unit and a control and communication unit, wherein the base platform is of a plane bearing structure and is provided with a through adsorption channel, the adsorption fixing unit is communicated with the base platform so as to form divisible adsorption fixation on the bearing surface of the base platform, and the visual acquisition unit vertically points to the bearing surface of the base platform with a fixed sight line and covers the full-width area of the bearing surface; the visual acquisition unit and the laser execution unit are electrically connected to exchange the shell appearance data, the cutting path set and the state data. According to the invention, the dependence on a fixing jig and manual intervention is reduced, and the cutting flexibility and automation degree are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of adaptive control, and in particular to an adjustable mobile phone shell laser cutting table. BACKGROUND

[0002] At present, the demand for intelligent manufacturing and personalized customization is growing, and the machining precision and appearance consistency of mobile phone shells, as important components of consumer electronics products, directly affect the overall quality. The existing mobile phone shell processing relies on numerical control milling, stamping forming or mold injection followed by local finishing. Although these methods are relatively mature in batch production, they have obvious shortcomings in small batch and multi-variety customization scenarios. For example, numerical control milling has multiple procedures and complex tool changing, making it difficult to flexibly adapt to the shape differences of different models of shells; stamping and injection are highly dependent on molds, which have long manufacturing cycles and high costs, making it difficult to meet the needs of rapid iteration and differentiated design. Therefore, how to ensure the shape precision while realizing flexible and adjustable shell processing has become a technical problem that the industry needs to solve.

[0003] In recent years, laser cutting equipment has been gradually applied to the processing of consumer electronics components. Laser cutting has the advantages of non-contact, high precision and adaptability to complex shapes, and is particularly suitable for hole cutting, edge trimming and complex contour cutting on shells. Some existing equipment positions the mobile phone shell through a fixed jig and then uses a laser scanning system to complete the cutting according to the preset trajectory. However, this type of solution generally relies on manual clamping and pre-made jigs. Each model of shell requires a specially designed jig, otherwise the positioning will not be accurate, the cutting path will not match the shell contour, and offset and waste products are likely to occur. For frequently changing shell designs, this method not only reduces production flexibility, but also increases tool consumption.

[0004] In addition, existing laser cutting platforms still have obvious limitations in shell recognition and path generation. Some solutions determine the shape contour of the shell through simple edge detection, but since the surface of the mobile phone shell is mostly curved, has a reflective coating or texture, edge detection is easily disturbed by noise, resulting in incomplete extraction results. Another type of solution attempts to use image template matching to identify the shell type, but this method usually requires a stable and clean background environment, and any slight obstruction, uneven lighting or shell position offset will cause identification failure. More importantly, the matching process of this type of solution is mostly rigid matching, lacking the ability to adapt to actual deformation and clamping deviations, making it difficult to ensure the accurate alignment of the trajectory and the shell. SUMMARY

[0005] The application aims to provide an adjustable mobile phone shell laser cutting table, which realizes the whole process of mobile phone shell from automatic identification, path generation to accurate cutting through the integrated cooperation of the base platform, the adsorption fixing unit, the visual acquisition unit, the laser execution unit and the control and communication unit. The core is to realize the high robustness identification of different models of shells by using concentric ring coding, topological fingerprint summary and skeleton symbol sequence image processing and matching methods, and to generate accurate cutting path through key point correspondence and piecewise linear deformation. Combined with the double gray level jump detection of cross neighborhood and the ghost segment elimination mechanism, the path can maintain continuity and reality under complex light and surface texture interference. During the cutting process, real-time correction is realized through fast re-encoding and local fine-tuning, so that the processing always maintains high precision. The application not only reduces the dependence on fixed jigs and manual intervention, significantly improves the flexibility and automation degree of cutting, but also has obvious advantages in processing precision, product consistency and production efficiency.

[0006] To solve the above technical problems, the application provides an adjustable mobile phone shell laser cutting table,

[0007] The adjustable mobile phone shell laser cutting table comprises a base platform, an adsorption fixing unit, a visual acquisition unit, a laser execution unit and a control and communication unit. The base platform is a plane bearing structure and is provided with a through adsorption channel. The adsorption fixing unit is in communication with the base platform to form a partitionable adsorption fixing on the bearing surface of the base platform. The visual acquisition unit is vertically directed to the bearing surface of the base platform with a fixed line of sight covering the full-width area of the bearing surface. The laser execution unit is arranged in a staggered manner relative to the visual acquisition unit with a fixed emission direction and the laser focal point falls on the bearing surface of the base platform. The control and communication unit is fixedly arranged in the base platform and is electrically connected with the adsorption fixing unit, the visual acquisition unit and the laser execution unit through a single wire bundle to exchange shell shape data, cutting path set and state data. The control and communication unit is internally provided with a shell template library and a path generator. After receiving the original image from the visual acquisition unit, the image preprocessing and contour extraction are sequentially performed, the anchor block retrieval and matching are performed according to the shell template library, the path generator is called to generate the cutting path set, the generated cutting path set is converted into trajectory instructions suitable for the laser execution unit and is issued, and the online micro-correction closed-loop control is implemented based on the state data returned from the visual acquisition unit.

[0008] Further, the mobile phone shell is placed on the bearing surface of the base platform and is stably attached by the adsorption fixing unit; the visual acquisition unit first acquires an empty reference image and then acquires a working image after the mobile phone shell is placed; the control and communication unit performs projection geometry setting on the working image based on the fixed mark points on the bearing surface to obtain a homographic mapping from the working image to the bearing surface coordinates; the acquired empty reference image is taken as the background, and the background subtraction, connected domain extraction and hole filling method are used to obtain a binary mask containing only the shell entity and a shell boundary pixel set.

[0009] Further, the control and communication unit performs successive erosion on the obtained binary mask until a single pixel or a very small pixel cluster first appears, records the arithmetic mean position of the row index and the column index of the pixel set formed and rounds it to the nearest pixel as the mask center point; takes the determined mask center point as the center, generates concentric rings in unit pixel radius increments until the shell boundary is covered; divides each concentric ring into one hundred and twenty sectors, and the sector number starts from a fixed reference direction in a clockwise direction; for each sector of each concentric ring, it is detected whether there is a shell boundary pixel; if there is a shell boundary pixel, it is recorded as occupied, and if there is no shell boundary pixel, it is recorded as empty; connect the one hundred and twenty occupied and empty states of a concentric ring into a fixed length ring code string according to the sector number sequence; perform the above operation on all concentric rings to obtain a group of ring code string sequences.

[0010] Further, the control and communication unit counts the number of occupied segments in each ring code string and the continuous length of each occupied segment, and records the start and end positions of each occupied segment according to the sector number; select three types of representative ring code strings from the generated entire ring code string sequence: the outer boundary representative ring code string with the longest total length of occupied segments, the inner ring representative ring code string with the smallest total length of occupied segments but not zero, and the middle ring representative ring code string with the total length of occupied segments in the middle; connect the position list and the length list of the occupied segments of the selected three representative ring code strings into a topological fingerprint digest according to the predetermined order.

[0011] Further, the control and communication unit performs thinning on the binary mask of the current mobile phone shell to obtain a skeleton pixel graph; select the boundary pixel with the maximum horizontal coordinate on the outer boundary of the shell after projection setting as the skeleton entry point; generate a skeleton walking track in an eight-neighborhood traversal manner; in the skeleton walking process, encode step by step based on the relative orientation of the current pixel and the last step pixel: if the next step pixel orientation remains unchanged from the last step, record it as a straight symbol, if the next step pixel is in the left neighborhood, record it as a left turn symbol, and if the next step pixel is in the right neighborhood, record it as a right turn symbol; when a bifurcation is encountered, press a branch stack in the order of the neighborhood where the bifurcation appears, and record the bifurcation entry symbol when entering the branch and the bifurcation exit symbol when returning to the trunk; record the termination symbol when reaching the endpoint; after completing the traversal, a skeleton symbol sequence covering the outer boundary and internal holes is obtained.

[0012] Further, the control and communication unit stores for each template in the candidate template set a skeleton symbol sequence of its template contour and a template key point list containing the symbol sequence number of the outer boundary starting point, each bifurcation position and each hole entrance position offline. Sequence alignment is performed between the generated current skeleton symbol sequence and the skeleton symbol sequence of each template, allowing only matching, skipping and merging of the two adjacent symbols into a single transition symbol as atomic operations. In the alignment, matching is preferred, followed by merging and then skipping. When there are multiple feasible alignment paths, the one with the least total number of atomic operations and the most number of matching is selected as the alignment result of the corresponding template. After the alignment of all candidate templates is completed, the one with the best score is selected as the final template. According to the key point list and the corresponding symbol sequence number in the alignment result of the determined final template, a one-to-one correspondence between the template key points and the current image key points is established. Linear deformation is performed within each triangle mesh formed by every three adjacent key points to map the template path sketch to the current image coordinates. Interpolation is used between the triangles on the shared edges to ensure path continuity.

[0013] Further, the control and communication unit performs boundary fine-tuning, specifically including: constructing a cross neighborhood perpendicular to the path tangent along each discrete point of the generated path sketch, taking a number of pixels on the inside and outside of the constructed neighborhood respectively, comparing the gray level jump amplitudes of the pixels on both sides, and selecting the pixel with the largest jump as the local boundary landing point. All local boundary landing points are connected in the original order to form a continuous path. At the corners, a sliding window of a fixed length is used to perform arithmetic averaging on the row and column indices of the path points to obtain a smooth corner. The generated continuous path is projected back to the working image to check whether each segment on the path satisfies the conditions of existence of gray level jump on both sides, distance from the nearest skeleton segment less than a preset pixel distance, and no self-intersection with other verified paths. Path segments that do not satisfy any condition are marked as ghost segments and deleted. If the deletion operation causes the path to break, a bridge segment is repeatedly generated in the neighborhood of the break endpoint centered at the break endpoint until the path is connected again.

[0014] Further, the control and communication unit sorts the continuous path that has passed the ghost segment test according to the outer boundary and each hole and specifies the cut-in and cut-out order to form the final cutting path set. The final cutting path set is converted from image coordinates to bearing surface coordinates using the obtained homographic mapping and stored as a trajectory instruction set.

[0015] Further, the control and communication unit downlink the generated trajectory instruction set to the laser execution unit for cutting in sections; the visual acquisition unit collects local images at fixed intervals, and the control and communication unit performs fast recoding of concentric rings and sectors in a small range around the current cutting position, and compares with the local ring code string of the corresponding template; when the number of locally unmatched sectors exceeds the preset threshold, only the local boundary fine-tuning is re-executed and the trajectory instruction set to be executed is updated; the adsorption fixing unit remains stable and adheres during the entire cutting process until all trajectory instruction sets are executed.

[0016] The adjustable mobile phone shell laser cutting table of the present application has the following advantages: through the formation of a highly integrated cooperative working mechanism between the base platform, the adsorption fixing unit, the visual acquisition unit, the laser execution unit and the control and communication unit, the processing of the mobile phone shell can automatically complete identification, path generation and accurate cutting without the need to change fixtures and without manual intervention. Compared with the existing technology which relies on fixed fixtures or simple edge detection, after the visual acquisition unit obtains the original image, the control and communication unit can not only use concentric ring coding, topological fingerprint summary and skeleton symbol sequence for robust identification of the shell shape, but also can realize one-to-one correspondence between the path and the actual shell by combining the template key point list, and ensure high-precision matching of different models of shells in deformation conditions through piecewise linear deformation. The boundary fine-tuning step detects the gray level jump of the cross neighborhood on both sides, so that the path landing point is more consistent with the real boundary, effectively reducing the error caused by reflection, gaps or texture interference. At the same time, the ghost segment inspection and bridging strategy ensure the continuity and reliability of the path, avoiding the generation of invalid cutting trajectories in complex texture environment. The final generated cutting path set is downlinked to the laser execution unit in the form of trajectory instruction set after sorting and coordinate conversion, and is checked in real time during the cutting process using fast recoding, and when local mismatch is detected, only the local fine-tuning and replacement are performed, ensuring the continuity and precision of the processing process. Through the above-mentioned manner, the present application not only improves the accuracy of shell identification and path generation, but also significantly enhances the robustness and adaptability in complex environment, reduces the dependence on fixtures and the degree of human intervention, so that the mobile phone shell laser cutting has higher flexibility, automation degree and product consistency, thereby improving the overall production efficiency and processing quality. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without creative labor.

[0018] Figure 1 The adjustable mobile phone shell laser cutting table provided by the embodiment of the present application has the advantages that DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme of the embodiment of the present application will be described clearly and completely below in combination with the drawings in the embodiment of the present application. Obviously, the described embodiment is a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiment in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] Reference Figure 1 The adjustable mobile phone shell laser cutting table comprises a base platform, an adsorption fixing unit, a visual acquisition unit, a laser execution unit and a control and communication unit. The adjustable mobile phone shell laser cutting table of the present application adopts integrated design, and each functional unit is coordinated to realize precise cutting processing. The base platform is a core bearing structure, which is configured in a horizontal rectangle, and the upper surface thereof forms a flat bearing surface for placing a mobile phone shell workpiece to be processed. A plurality of adsorption channels are provided through the base platform, and are uniformly distributed in the form of black rectangular holes, so as to ensure effective adsorption and fixation of different sizes of shells. The visual acquisition unit and the laser execution unit are respectively located on the left and right sides above the base platform, and are arranged in a symmetrical staggered manner. The visual acquisition unit is fixedly directed to the bearing surface through a vertical line of sight, and the coverage range thereof is indicated by a dashed line boundary, so as to realize full-width monitoring of the entire bearing surface. The laser execution unit is installed in a staggered manner relative to the visual unit, and a laser beam is indicated by a black solid line, which is accurately focused on a predetermined cutting position on the bearing surface to form a laser focal point. The control and communication unit is fixedly arranged in the base platform, and adopts modular design, and comprises two core functional modules of a shell template library and a path generator. The unit is electrically connected with each execution unit through a single wire system, so as to realize unified transmission of image data, cutting instructions and state feedback. The adsorption fixing unit is independently arranged, is connected with the adsorption channels in the base platform through a connecting pipeline, and is uniformly dispatched by the control unit to implement differential adsorption force control on different regions, so as to ensure stable positioning of the shell workpiece during cutting. The whole system forms a closed-loop control system, the visual unit acquires workpiece image information in real time, the control unit performs matching identification and generates a cutting path based on the template library, the laser unit performs accurate cutting according to the generated trajectory instructions, and online correction is realized through visual feedback to ensure processing precision and consistency.

[0021] The base platform is a monolithic planar load-bearing structure, employing a combination of high-rigidity sheet metal and honeycomb sandwich panels, with a load-bearing surface machined on its upper surface. A continuous adsorption channel is arranged within the load-bearing surface, connecting to the adsorption and fixing unit via an internal confluence cavity. Several mechanical reference points and fixed markers for the projection geometry calibration of the vision acquisition unit are set on the load-bearing surface. The choice of a monolithic planar load-bearing structure with an internal confluence cavity provides uniform adsorption without increasing the overall dimensions, thus maintaining stable adhesion even when the phone casing has local undulations or edge warping, reducing deformation errors observed by the vision acquisition unit and focal plane drift of the laser actuation unit.

[0022] The adsorption and fixation unit includes a suction source, a zoned control valve assembly, and a return filter component. It communicates with and is controlled by a communication unit via electrical connections in a single wiring harness. Below the bearing surface, several zoned adsorption channels are divided into a grid. Each zone is connected to a corresponding control valve assembly, enabling zoned adsorption and fixation. The advantages of zoned adsorption are: when the phone casing has camera openings, logo recesses, or localized damage, the uncovered adsorption holes do not significantly reduce the overall negative pressure level, thus ensuring stable adhesion; simultaneously, zones closer to the edge can be used for thin or warped areas, improving edge holding capacity and reducing edge drift compensation during subsequent path generation.

[0023] The visual acquisition unit is fixedly mounted above the base platform, pointing vertically towards the bearing surface with a fixed line of sight and covering the entire image area. The visual acquisition unit includes a lens assembly, image sensor, uniform area illumination, and a polarizing anti-reflective layer. The vertical line of sight and full-frame coverage facilitate mapping the working image to the bearing surface coordinates. The use of uniform area illumination in conjunction with the polarizing anti-reflective layer significantly reduces strong reflection spots caused by the coating and protective film on the housing surface, resulting in more stable image preprocessing and contour extraction. When the housing surface is dark or textured, the combination of polarizing and uniform illumination reduces the proportion of overexposed and underexposed areas, improving boundary integrity.

[0024] The laser actuator is positioned offset from the vision acquisition unit with a fixed emission direction to ensure the laser focus falls on the bearing surface. The laser actuator includes an energy source, a two-axis beam guide, a focusing assembly, and an air curtain protection structure. The two-axis beam guide performs angular scanning along the trajectory command, the focusing assembly converges the laser spot to the vicinity of the bearing surface, and the air curtain protection structure coaxially sprays clean airflow along the emission direction to isolate molten slag and fumes from contaminating the mirror surface and the emission port. This fixed emission direction and offset arrangement from the vision acquisition unit prevents them from obstructing each other; simultaneously, the fixed emission direction facilitates the use of a single geometric calibration in the control and communication units, directly associating the trajectory command with the bearing surface coordinates and reducing intermediate conversion errors.

[0025] The control and communication unit is fixed inside the base platform and electrically connected to the adsorption fixing unit, vision acquisition unit, and laser execution unit via a single wiring harness. This enables the exchange of shell shape data, cutting path sets, and status data. The control and communication unit incorporates a shell template library and a path generator. Upon receiving the original image, it sequentially performs image preprocessing and contour extraction, anchor block retrieval and matching based on the shell template library, generates a cutting path set using the path generator, converts the cutting path set into trajectory commands and issues them, and performs closed-loop control with online micro-correction based on status data. Centralizing these functions within the control and communication unit reduces the number of connections, minimizes signal interference, and ensures a closed-loop image-to-track mapping link within the same processing environment, facilitating unified management of time synchronization and error sources. The single wiring harness includes power pairs, data pairs, and a shielding layer, employing a layered braided shielding and independent grounding structure to ensure that the high-speed data from the vision acquisition unit and the drive signals from the laser execution unit do not interfere with each other during co-transmission. Using a single wiring harness reduces the number of external connection points, decreases status data jitter caused by poor contact, and improves the reliability of closed-loop control.

[0026] Furthermore, place the phone casing flat on the base platform with the outer side facing up. First, activate the adsorption channels near the edge of the casing, then activate the central adsorption channel. This allows the adsorption unit to form a ring-shaped pressure at the outer edge before suctioning the center. This sequence first constrains the circumferential freedom of the casing, then eliminates bulges or slight warping in the center, reducing boundary drift and deformation in subsequent images. After initiating adsorption, hold the casing still for at least 2 seconds to allow the material's slight elastic recovery process to complete. The purpose of this stillness is to stabilize the contact state between the casing and the support surface, avoiding differential noise caused by slow deformation between two images.

[0027] The visual acquisition unit points vertically towards the support surface with a fixed line of sight and acquires an empty reference image with consistent exposure and illumination. To reduce random noise, three consecutive shots are taken, and the pixel-wise average is used as the final empty reference image. Averaging suppresses random fluctuations in light sensitivity and micro-flicker in illumination, resulting in cleaner background subtraction. After placement and alignment with the housing, a working image is acquired without changing the lens focal length, focus, exposure, or illumination. Maintaining consistent acquisition conditions ensures that the empty reference image and the working image are comparable in brightness and contrast, with differences primarily originating from the housing itself, thus making background subtraction more discriminative. If there is slight reflection at fixed marker points on the support surface, the highlights are reduced by lowering the illumination angle or using a polarizer without changing the exposure; this preserves the geometric outline of the marker points without sacrificing detail in the shadows, facilitating subsequent projection adjustment.

[0028] At least four fixed marker points are pre-arranged on the bearing surface of the base platform, preferably at the four corners of a rectangle with an additional auxiliary marker point at the midpoint of each side. The four corners of the rectangle provide a stable full-frame geometric reference, and the auxiliary marker points are used to ensure calibration even when corner points are occluded or locally contaminated. The fixed marker points are located in both the unloaded reference image and the working image. The location process is as follows: first, threshold segmentation is performed across the entire image to obtain candidate high-contrast regions; then, non-marker regions are filtered by area and aspect ratio; finally, the centroid pixel is calculated within the candidate regions as the marker pixel position. This dual filtering by area and aspect ratio eliminates adsorption channel openings and occasional contaminants. Using the unloaded reference image as a reference, the four corner marker points are sorted in the order of "top, right, bottom, left"; in the working image, they are sorted in the same order and correspond one-to-one with the unloaded reference image. The advantage of using a fixed order is to prevent geometric inversion caused by left-right or top-bottom swapping. The homography mapping from the working image to the bearing surface coordinates is determined using the corresponding four corner marker points in the two images. To reduce the impact of lens edge distortion on calibration, if auxiliary markers exist, the four corner markers can be used as the main reference, combined with two auxiliary markers near the image center for consistency verification. When the mapping given by the correspondence between the four corners and the auxiliary markers shows higher consistency in the central region, the solution including the auxiliary markers should be prioritized. This approach improves the geometric accuracy of the mapping to the central processing area, ensuring accurate alignment of the cutting path set with the bearing surface coordinates in the critical region. The working image is then calibrated to the bearing surface coordinates using homography mapping to obtain the calibrated working image. In the calibrated image, the lines connecting the straight edges of the bearing surface to the fixed markers should appear as horizontal or vertical pixel rows and columns, which is beneficial for subsequent pixel-level scanning and connected component determination.

[0029] Pixel alignment is checked between the unloaded reference image and the calibrated working image: the average brightness difference is calculated within a small neighborhood of the fixed marker point. If the average value exceeds the predetermined tolerance, it indicates exposure drift between the two images. White balance and brightness should be unified first to make the brightness difference between the two images in the neighborhood of the marker point as close to zero as possible. Unifying brightness has the advantage of approximately canceling out the background portion after differencing, retaining only the structural differences introduced by the shell. The difference image is obtained by subtracting the unloaded reference image from the calibrated working image. In the difference image, the shell-covered area exhibits significant brightness or texture differences due to obscuring the original surface texture, while the difference in the background area is close to zero.

[0030] Connected component extraction is performed on the initial binary image, using 8-neighborhood to determine pixel connectivity. The reason for choosing 8-neighborhood over 4-neighborhood is that the outer shell edges often appear as diagonal pixel chains, and 8-neighborhood maintains this diagonal continuity, preventing the continuous boundary from being mistakenly broken into multiple segments. For each connected component, three metrics are calculated: area, the length and width of its circumscribed rectangle, and the average distance from the fixed marker. The outer shell body typically corresponds to connected components with a larger area, whose circumscribed rectangle's length and width match the shell's shape, and which are far from the fixed marker. These three metrics are used to rank the connected components, and the top-ranked component is selected as the outer shell body. This selection ensures stable identification of the outer shell region even with interference from tool shadows or localized paper scraps. All small connected components outside the outer shell body are removed, resulting in a binary image containing only the outer shell body. To enhance the separability of the outer shell boundaries, the difference image is smoothed to suppress isolated noise points before thresholding to obtain the initial binary image. Smoothing first aims to eliminate the interference of individual bright spots on thresholding decisions, making thresholding more stable.

[0031] In a binary image containing only the outer shell, internal holes are identified. Sources of these holes include camera openings, microphone openings, logo recesses, and localized breaks caused by reflections. For each hole, its area and roundness are assessed. Holes with excessively small areas and poor roundness are often noise-induced pseudo-holes and should be directly filled; holes with areas and roundness matching known hole locations and shapes are retained. This determination preserves the true hole locations while repairing noise-induced holes at the mask level. A closure repair is performed on the outer shell's edges: small gaps are bridged along the edge direction to connect adjacent narrow seams, ensuring a continuous outer shell contour. The significance of bridging is to allow subsequent outer shell boundary pixel sets to form closed loops, facilitating sorting and tracking.

[0032] Extracting shell boundary pixels from the repaired binary image: Scan the entire image and check if there are background pixels in the 8-neighborhood of each foreground pixel; if so, the foreground pixel belongs to the shell boundary pixel. This neighborhood difference judgment method can accurately identify the boundary position at the pixel level, avoiding dependence on gradient threshold. Organize all shell boundary pixels into an ordered set in a coherent order. First, search for the first boundary pixel in the upper left region as the starting point, and then move forward step by step along the adjacent boundary pixels, prioritizing candidate pixels that are consistent with the current direction or have the smallest turning angle to achieve single-loop tracking. When returning to the starting point, the complete sequence of the outer boundary is obtained. Repeat the same tracking for each internal hole to obtain the boundary sequence of each hole position. The direction-first tracking strategy helps to maintain the overall order stability when there are jagged edges or local burrs on the boundary, and avoids jumps and loops. Perform a consistency check on the ordered set of shell boundary pixels: Calculate the minimum distance distribution of the line connecting the boundary pixels to the fixed marker point. If a very small number of abnormally high points appear, usually caused by isolated noise, they can be averaged with the positions of the two preceding and following pixels to smooth them out. This eliminates tiny spikes on the boundary without introducing formulas, making the boundary more suitable for subsequent path generation and projection comparison.

[0033] Furthermore, the control and communication unit performs successive erosion on the binary mask until a single pixel or a very small pixel cluster first appears. The significance of successive erosion is to peel away edge details and local gaps from the outside in, allowing the remaining area to naturally converge to the vicinity of the shape's center. Compared to one-time geometric fitting, this process is less sensitive to mask gaps caused by edge damage or local reflections. The row and column indices of the final retained pixel set are recorded, and the arithmetic mean of the row and column indices is calculated, then rounded to the nearest pixel to form the mask center point. The advantage of rounding continuous values ​​to the pixel point is that subsequent concentric rings and sector divisions are performed on the pixel grid. If the center point is aligned with the grid, the pixel assignments at the ring and sector boundaries will remain consistent, preventing shifting between adjacent pixels.

[0034] Concentric rings are generated with the mask center point as the center, increasing by unit pixel radius. The choice of unit pixel step size ensures that there are no uncoded radius gaps between adjacent rings, facilitating subsequent statistics on the occupancy distribution of the shell boundary at different radii. The stopping condition for "until the shell boundary is covered" is determined by traversing the set of shell boundary pixels, calculating the pixel distance from each boundary pixel to the mask center point, and taking the maximum pixel distance as the radius of the outermost concentric ring. When the radius of the generated outermost concentric ring is not less than this maximum pixel distance, the concentric ring is considered to have covered the shell boundary. This method of determining the outermost ring ensures that all boundary pixels fall within the radius of a certain ring, without omitting protrusions or extensions far from the center.

[0035] Each concentric ring is divided into 120 equal sectors, with sector numbering starting clockwise from a fixed reference direction. The reference direction is the horizontal orientation of the adjusted image, i.e., from left to right. The line connecting the edge and the marker point is aligned with both the horizontal and vertical directions. Using this direction as a reference helps maintain consistent numbering across different batches of equipment. The consideration for dividing the sectors into 120 equal parts is that this resolution can subdivide the angular differences of common shell rounded corners and multiple openings, avoiding both excessive coarseness that could cause confusion in the angular dimension of different shapes, and excessive fineness that would result in too few effective pixels in a single sector, increasing its susceptibility to noise. If subsequent processes require higher angular subdivision, the number of sectors can be adjusted in optional implementations, but in this embodiment, it is fixed at 120.

[0036] For each concentric ring, an occupancy marker table of length 120 is established. Initially, all 120 positions are set to empty. The set of outer shell boundary pixels is traversed, and for each boundary pixel, a two-step classification process is performed: based on the pixel distance from the boundary pixel to the mask center point, the distance is rounded to the radius of the nearest concentric ring, and the boundary pixel is assigned to the corresponding concentric ring. Rounding is used to absorb sub-pixel level rounding errors and avoid boundary pixels jittering between two adjacent rings. The orientation angle of the boundary pixel relative to the mask center point is calculated and mapped to a specific number in one of the 120 equally divided sectors. The orientation angle is derived from the relative relationship between the row and column indices. Using orientation angle quantization instead of pixel-by-pixel ray scanning maintains the consistency of angle allocation without changing the image acquisition conditions. In the occupancy marker table of the corresponding concentric ring, the sector position obtained from the orientation classification is marked as occupied. If multiple boundary pixels exist within the same sector, only the occupied status is retained. Once all boundary pixels of a concentric ring have been classified, the occupancy markers of that ring are concatenated in sector number order to form a ring code string of fixed length. Each position in the ring code string has only two states: occupied or empty. Repeating the above process for all concentric rings yields a sequence of ring code strings arranged from the inside out. This sequence decomposes the complex two-dimensional geometric contour into multiple one-dimensional occupancy sequences, facilitating subsequent statistical analysis of structural features at different radius levels. Since the center point, concentric rings, and sectors are all defined on the pixel grid, the ring code string is stable against slight changes in illumination and local reflections.

[0037] In real-world images, the outer shell boundary may exhibit extremely short discontinuities. To avoid misjudging these "slits" caused by reflections or localized underexposure as empty, a proximity consistency check can be performed after the ring code string is generated: when a single empty sector is sandwiched between two occupied sectors and there are continuous boundary pixels near the radius of the concentric ring corresponding to the empty sector, the empty sector can be considered occupied. This process is equivalent to bridging extremely short gaps, making the ring code string closer to the coherence of the real boundary, while avoiding incorrectly filling longer gaps or real notches, because longer gaps often span multiple consecutive empty sectors at the sector scale, failing to meet the condition of "single empty sector".

[0038] For each ring code string, perform a clockwise scan, counting the number of occupied segments and the continuous length of each occupied segment. An occupied segment is defined as the largest continuous segment consisting of several adjacent occupied sectors; when an occupied segment crosses the boundary between the end and start of the ring code string, it is treated as a continuous occupied segment to avoid artificial breakage. For each occupied segment, record its start and end positions in the sector number. The reason for using start and end positions instead of just recording the length is that the position carries angle information, which can be directly used for subsequent alignment with the template without having to re-locate the angle on the original image.

[0039] Outer boundary representative ring code string: Among all ring code strings, calculate the total length of the occupied segment for each ring code string (i.e., the sum of all occupied segment lengths). Select the ring code string with the largest total occupied segment length as the outer boundary representative ring code string. The basis for this selection is that rings closer to the outer boundary of the shell usually cover longer continuous occupied segments, which can stably depict the overall outline of the shape. If there are ties, prioritize the ring code string with a larger radius to ensure that it represents the shape information of the outermost layer. Inner ring representative ring code string: Among the remaining ring code strings, select the ring code string with the smallest but non-zero total occupied segment length as the inner ring representative ring code string. This ring is usually located near the center point of the mask, corresponding to the general shape inside the shell and the overall distribution of smaller holes, and can provide an indication of the presence and sparseness of the internal structure. If there are ties, prioritize the ring code string with a smaller radius to make it closer to the central area. Middle ring representative ring code string: After eliminating the first two types of representatives, sort them according to the total occupied segment length from smallest to largest, and select the one located at the median position as the middle ring representative ring code string. The middle ring, representing the ring code string, is often located in the transition area of ​​the shape. It can compensate for the difference in angular occupation between the outer boundary ring code string and the inner ring code string, enabling the fingerprint to have discrimination ability at three scales simultaneously.

[0040] The outer boundary representative ring code string, the inner ring representative ring code string, and the middle ring representative ring code string are each used to retrieve their respective occupied segment position lists and occupied segment length lists, and then concatenated in a fixed order to form a topological fingerprint digest. The fixed order is: outer boundary representative ring code string first, inner ring representative ring code string in the middle, and middle ring representative ring code string last. The use of three representative ring code strings instead of all rings is based on a balance between information density and stability. The outer boundary representative ring code string carries the overall shape outline, the inner ring representative ring code string reflects the existence and sparse characteristics of the internal structure, and the middle ring representative ring code string captures the angular distribution of the transition layer. This combination can cover most of the geometric discrimination information without increasing the data volume excessively. This topological fingerprint digest is stored in the control and communication unit for subsequent rapid retrieval and matching with the shell template library. Since the sector numbering and reference direction are unified after tuning, the fingerprint digest can be stably reused across devices and batches.

[0041] The positions of the occupied segments representing the outer boundary ring code string are mapped back to the projected image coordinates using homography. It is then checked whether the corresponding angular intervals all have outer shell boundary pixels falling on concentric rings close to the outer boundary. If certain angular intervals are found to be missing for extended periods, the process returns to the ring code string generation step to check if any consecutive empty sectors near that radius are incorrectly preserved. When severe damage or occlusion of the outer shell boundary causes large gaps in the ring code string, the outermost ring with the larger total length of its occupied segments can be used as a replacement for the outer boundary ring code string. This avoids the selection of a representative ring due to abnormal noise in a single ring.

[0042] In addition to successive erosion, a distance map from pixels to boundaries can be calculated for the binary mask. The pixel with the largest distance value is selected as the center candidate, and then the arithmetic mean of the row and column indices is calculated within a small neighborhood centered on this candidate and rounded to obtain the mask center point. This approach is more robust when there are large gaps at the shell edge. When the resolution of the vision acquisition unit is significantly improved and the boundary pixel density is higher, the number of sectors for each concentric ring can be set to 180 to obtain higher angular discrimination capability; when pursuing higher processing speed and large shape differences, it can be set to 96 to obtain faster encoding and comparison. Regardless of the number used, the numbering order and reference direction remain unchanged to ensure consistency across devices. In sector occupancy detection, in addition to "marking an occupancy as the presence of any shell boundary pixel", the condition "marking an occupancy only when at least two adjacent boundary pixels appear" can also be used to suppress false judgments caused by a single isolated pixel; when only a single isolated pixel appears, it is left empty. This strategy is suitable for cases with strong surface texture or a lot of occasional noise. For proximity consistency checks, a bridging length of two adjacent empty sectors is allowed to handle gaps caused by boundaries in low-contrast regions. However, it is not recommended to exceed two to avoid misjudging genuine notches as continuous boundaries. When the outer boundary represents parallel ring code strings, in addition to prioritizing rings with larger radii, the average continuous length of occupied segments can also be compared, with rings having larger average continuous lengths being preferred to highlight the integrity of the overall outline. When the inner ring represents parallel ring code strings, in addition to prioritizing rings with smaller radii, the number of occupied segments can also be compared, with rings having fewer occupied segments being preferred to emphasize the sparsity of the internal structure.

[0043] The control and communication unit shrinks the binary mask into thin, connected lines of single-pixel width while maintaining the connectivity and endpoint positions of the main body of the shell, avoiding the generation of isolated pixels. This folds the width information into the central connection line, and subsequent traversals only need to follow the skeleton pixel map to cover the topological relationship of the shape and hole positions, significantly reducing the computational load and making it insensitive to local brightness fluctuations. The foreground pixels are scanned across the entire image from top to bottom and left to right, identifying pixels that meet the criteria of being "located on the boundary, not disrupting connectivity after deletion, and not an endpoint," and marking them for deletion. Marked pixels are deleted uniformly after each scan round; this process is repeated until no new markings are added in a particular scan round. This batch marking and unified deletion method avoids cascading effects within the same round, ensuring the reliability of connectivity judgment. Pixels connected to only one foreground neighbor are considered endpoints and are not deleted in any round. When a short branch with a length of 3 pixels or less is detected, the entire branch is deleted to prevent burrs caused by reflections or minor scratches from persisting in the skeleton and affecting the stability of subsequent branch judgments and symbol encoding. If no pixels are marked as to be deleted after two consecutive scans, this is considered skeleton convergence, and the skeleton pixel map is output. This convergence criterion facilitates obtaining a consistent termination time across different batches of images.

[0044] In the set of pixels at the outer shell boundary, the closed boundary curve with the largest circumscribed rectangle area is selected as the outer boundary. The largest circumscribed rectangle area usually corresponds to the outer contour of the phone shell, which can be distinguished from the internal hole boundaries. On the outer boundary, the boundary pixel with the largest lateral coordinate after projection adjustment is found as the skeleton entry point. The reason for selecting the pixel with the largest lateral coordinate is that this position is not sensitive to small changes in the geometric relationship between the camera and the supporting surface, and it is located near the right edge of the image in most placement postures. The entry point selection has stable reproducibility, which is convenient for cross-batch comparison and debugging. For each skeleton pixel, its eight neighbor position numbers are recorded in a clockwise direction. The numbers are consistent with the row and column direction of the image to ensure that the traversal order is consistent across devices. Starting from the skeleton entry point, the current direction is set to a leftward horizontal direction (consistent with the largest lateral coordinate). At each step, the neighboring pixels consistent with the current direction are selected as candidates for the next step. If there are no candidates with the same direction, the neighboring pixel with the smallest angle with the current direction is selected in a clockwise order. If multiple candidates appear, the one with the larger lateral coordinate is selected to continue the directional consistency at the beginning. When there are two or more valid candidates at a certain position and there is no option to return to the previous pixel, push all other candidates except the optimal candidate onto the branch stack in clockwise order of neighborhood, and record the current position as the entry point of these branches; the traversal continues along the optimal candidate. When the current path reaches a point where there are no feasible candidates or reaches an endpoint, pop the most recently pushed branch entry point from the branch stack and switch to that branch to continue traversing. This depth-first strategy ensures that the entire skeleton is covered without backtracking a large range of paths. Each skeleton pixel is marked as visited the first time it is visited. If a visited pixel is encountered again and is not the source of the previous traversal, it is considered a loop, and the next candidate is selected according to the rules to avoid infinite loops in closed structures. The traversal ends when the branch stack is empty and there is no feasible next step in the current path. At this time, all skeleton pixels should have been visited at least once; if there are unvisited pixels, it indicates that there are broken or isolated branches, and it is necessary to return to the thinning and spur suppression steps for verification.

[0045] Based on the relative position of the current pixel and the previous pixel, if the next pixel maintains the same position as the previous pixel, a straight-ahead symbol is recorded; if it is in the left neighbor, a left-turn symbol is recorded; and if it is in the right neighbor, a right-turn symbol is recorded. These three types of symbols are sufficient to express directional changes within eight neighbors and can cover the actual walking trajectory with a minimal set. When a candidate path is pushed onto the branch stack, a fork-in symbol is recorded. When returning to the entry point and switching to the pushed branch, the fork-in symbol is not recorded again. When a branch reaches its end and returns to the entry point to continue to the next branch, a fork-out symbol is recorded. Fork-in and fork-out symbols allow the branch structure to be directly reconstructed at the symbol level without having to look back at pixel-level coordinates. When a pixel with only one neighbor is reached and that neighbor is the source pixel, a termination symbol is recorded. For closed loops, termination symbols are not recorded; instead, the loop is closed and traversal continues until the branch stack is exhausted. After the complete traversal, the skeleton symbol sequence should cover the structure of the outer boundary and internal holes, containing an ordered set of straight-ahead symbols, left-turn symbols, right-turn symbols, fork-in symbols, fork-out symbols, and termination symbols. The symbol sequence is divided into fixed-length windows. The proportion of straight symbols in the window is checked to see if it matches the straight segments of the corresponding line. If the symbols in the window frequently alternate left and right and correspond to short jagged edges on the pixel map, the process can be returned to the refinement stage to increase the strictness of spur suppression in order to obtain a smoother skeleton walking trajectory.

[0046] The control and communication unit simultaneously advances from the beginning of both sequences, judging the current symbol pair: if the symbols on both sides are the same, matching is performed, and each sequence advances one step simultaneously; if they are different, but the combination of the current symbol on one side and its next symbol on the other side forms a smoother turning trend at the corresponding position, merging is performed on that side, treating the adjacent two symbols as a single transition symbol before attempting matching again; if neither matching nor suitable for merging, the side most likely to be noise is skipped, the other side remains unchanged, and the judgment is re-evaluated. When multiple feasible alignment paths exist, the total number of atomic operations and the number of matches for each path are recorded. First, they are sorted by the total number of atomic operations from smallest to largest, then by the number of matches from largest to smallest, and the path with the highest ranking is selected as the alignment result for that template. If a tie still occurs, the path where the first uncertain branch occurs later in the alignment process is selected, making the two sequences completely consistent over a longer prefix, thereby improving global consistency. The above alignment and scoring process is repeated for each template in the candidate template set, and the one with the best score is selected as the final template.

[0047] Based on the symbol sequence number in the final template keypoint list, locate the corresponding current skeleton symbol sequence position in the alignment results, establishing a one-to-one correspondence between template keypoints and current image keypoints. If the corresponding position of a template keypoint is missing in the current sequence (e.g., a broken skeleton at the entrance of a hole), search for a fixed number of symbols along the sequence before and after that position, selecting the point with higher consistency with the template symbol context as a replacement. If there is a one-to-many situation with template keypoints (e.g., two approximate paths at a fork), select the side with higher pixel density at the shell boundary as the correspondence to reduce deviations after projection deformation. Project the established keypoint pairs onto the working image and check if the connection is consistent with the main direction of the skeleton pixel map. If there is a significant reversal in a local area, it indicates that the sequence alignment has skipped inappropriately at that point, and the alignment step should be returned to prioritize merging in that local area for a smooth transition.

[0048] Triangles are generated in groups of three adjacent keypoints, and triangulation networks are constructed according to the outer boundary ring and each hole location ring. In areas with bifurcation, local triangles are preferentially formed by the bifurcation entry point, bifurcation exit point, and adjacent inflection points to improve adaptability to local deformation. The coverage of each triangle should be non-overlapping and fill the outline of the shell body as much as possible. The template path sketch is superimposed on the template triangulation network, recording which template triangle each path point is located in or on which shared edge. For a path point located inside a template triangle, the proportional relationship of the point relative to the three vertices of that triangle is kept unchanged, and it is placed in the same proportional position inside the corresponding triangle in the current image. For a path point located on a shared edge, it is directly located in the same proportional position on the shared edge in the current image along the corresponding mapping of the shared edge. The scheme of maintaining the relative proportion can achieve linear interpolation of shape without introducing complex computational expressions, which can ensure the continuity of local deformation and keep the overall topology unchanged. On the shared edges of adjacent triangles, the same set of edge mapping results is used for both sides of the triangle, and the calculation is not repeated, thereby avoiding cracks or ghosting at the boundary. Within each triangle, the traversal direction is determined according to the order of key points in the symbol sequence, maintaining the vertex order of the template triangle and the current triangle consistent. When a local triangle is detected to have an opposite direction to its neighboring triangles, the vertex order of the neighboring triangles is adjusted first to restore consistency. Consistent direction prevents local mirroring or folding after mapping. After mapping all path points, a copy of the path sketch located in the current image coordinates is obtained, which is continuously connected between the triangles and can directly proceed to subsequent boundary fitting fine-tuning and phantom segment removal.

[0049] In addition to "prioritizing the current direction," "right-hand rule priority" or "left-hand rule priority" can also be used as alternatives. As long as the same rule is used in offline template generation and current movement, the stability of the symbol sequence alignment can be maintained. When the resolution of the visual acquisition unit is high and the edge noise is more fragmented, "delete short branches with a length of less than or equal to 3 pixels" can be adjusted to "length less than or equal to 5 pixels"; in low-resolution scenes, it can be kept at 3 pixels to avoid accidentally deleting real small hole entrances. When multiple branches appear consecutively at the same position within a short distance, the entry sequence number of the branch can be attached along with the bifurcation entry symbol, and the same rule can be maintained on the template side; this approach can improve the alignment discrimination in areas with dense hole positions.

[0050] Discrete points are selected along the path sketch generated by the mapping with a fixed step size. Ideally, each adjacent point should be 2 to 3 pixels apart on the image to ensure that subtle transitions are captured without excessive redundancy. For rounded corners and narrow areas, the step size is temporarily reduced to 1 pixel to improve positioning accuracy. At each discrete point, the tangent direction of the approximate path is determined based on the line connecting that point to its immediate vicinity and the point before and after it. The direction perpendicular to this tangent is then taken as the normal. Centered on the discrete point, pixel lines of equal length are extended along both the normal and the tangent directions to form a cross-shaped neighborhood. Ideally, each line has a radius of 5 pixels and a total length of 11 pixels. The reason for using the cross-shaped neighborhood is that the normal is used to accurately cross the boundary, and the tangent is used to determine whether the local texture extends along the boundary direction, thus distinguishing the real boundary from the texture extending along the tangent.

[0051] On the normal line of the cross neighborhood, take 5 pixels inward and outward from the center point, calculate the grayscale change amplitude within each range, and record the maximum jump position on both sides. Select the position with the larger amplitude of the maximum jump on both sides as the local boundary landing point of the sampling point. The reason for comparing both sides is that the boundary of real material usually exhibits a sudden change in reflection or transmission properties from the inside to the outside. Simultaneous scanning on both sides can eliminate false jumps caused by shadows on one side; selecting the position with a larger amplitude helps to ensure that it still falls on the real boundary when reflections or blemishes are present. Connect all local boundary landing points into a broken line according to the original order of the discrete points. For individual positions with missing landing points (e.g., local strong reflections causing no effective jumps to be detected on either side), replace them with the line connecting the two nearest determined landing points to avoid creating breakpoints. At corners, use a fixed-length sliding window to perform an arithmetic mean of the row and column indices of the path points. The preferred window length is 7 points, with the window center aligned with the path points near the corner. This approach eliminates high-frequency jagged edges without changing the overall geometry, making the movement of the laser actuator continuous and reducing pauses and overshoot.

[0052] The smoothed continuous paths are directly verified pixel-by-pixel on the working image. Each segment is checked for the following three conditions, all of which must be met simultaneously: **Existence of grayscale transitions on both sides:** Along the normal of each path point, there are obvious grayscale transitions on both sides. This confirms that the path point indeed crosses the boundary between the material and the background, rather than falling within the texture. **Distance to the nearest skeleton segment is less than a preset pixel distance:** Using the skeleton pixel map as a reference, find the nearest skeleton pixel, preferably less than or equal to 3 pixels. The skeleton represents the centerline network of the shape; a path that is moderately close to the skeleton indicates that the path follows the overall geometry; excessive distance often means crossing into reflective or shadow areas unrelated to the subject. **No self-intersection with other verified paths:** Verified path segments are kept free of self-intersection constraints to prevent path crossings in densely populated hole areas, which could disrupt the subsequent cutting order. Path segments that do not meet any of the conditions are marked as phantom segments and deleted. Deletion prioritizes whole-segment deletion: A set of consecutive unmet path points is removed as a whole. Whole-segment deletion avoids unnecessary jitter caused by residual fragments in subsequent sorting and execution.

[0053] If deleting a phantom segment causes a path break, a cross-neighborhood search is reapplied within the neighborhood of the break endpoints, advancing 3 to 5 sampling steps along the tangential directions of both ends to generate two short candidate bridging segments. Bilateral grayscale jump checks and skeleton distance checks are then performed on these candidate bridging segments again, retaining only those that simultaneously meet the criteria, and connecting the pair with the smallest error at both ends. This strategy prioritizes bridging along the original geometric trend, avoiding detours in complex textures. Verification is considered complete when no new phantom segments exist on the path and all bridging segments pass the bilateral jump and skeleton distance checks. For cases with multiple holes, the above verification process is performed separately for the outer boundary and each hole.

[0054] The validated continuous paths are sorted by outer boundary and each hole position. The priority order is: holes first, then outer boundary. Inside the hole position, the process proceeds from the center area to the edge area; when adjacent holes are very close, they are processed continuously within a group to reduce back-and-forth processing. The hole-first-outer-second order allows the adsorption and fixing unit to maintain the maximum effective contact area for most of the time, reducing pose changes caused by the opening of the outer contour. An entry point and an exit point are set for each path. The entry point is preferably on a straight segment of the path, with the normal pointing towards the area where the adsorption channels are denser; this provides more stable material support at the moment of entry, reducing entry burrs caused by minor warping. The exit point is preferably offset from the entry point by at least 20 degrees to avoid the heat-affected zone overlapping at the same angle. Homography is used to transform the cutting path set from image coordinates to bearing surface coordinates. A sparser sampling density is used for long straight segments to improve execution speed; a denser sampling density is used for small radius corners and hole edges to ensure contour accuracy. The change in sampling density is connected at the path connection point by overlapping 1 to 2 points to ensure a continuous transition of the laser actuation unit.

[0055] Each sorted path is divided into multiple segments, each containing a set of sequentially arranged control points and an execution direction marker, seamlessly connected to each other. The segment length is preferably between 8 and 20 control points to balance execution continuity with the flexibility of online correction insertion. This results in a trajectory instruction set, organized according to both path and segment order. The control and communication unit sends the trajectory instruction set segment by segment to the laser execution unit for execution. After each segment is completed or after a fixed time interval (e.g., 50 milliseconds, whichever comes first), the vision acquisition unit acquires local images around the current cutting position and transmits status data back. Using segments as the unit of advancement allows for stable alignment of processing actions and perception verification without interrupting the continuous movement of the laser execution unit. A local region with a radius of 40 to 60 pixels is selected centered on the current cutting position, and only concentric rings and sectors within this region are rapidly recoded to obtain a local ring code string. This local ring code string is compared sector by sector with the corresponding local ring code string of the template, and the number of mismatched sectors is counted. The reason for using local ring code strings is that rings and sectors are naturally robust to slight scale and illumination changes, and the computational cost is much lower than full-width rematching, making them suitable for high-frequency operation during execution. When the number of locally mismatched sectors exceeds a preset threshold (preferably 4 to 6), boundary fine-tuning and continuous path verification are re-executed only within that local area, and the corresponding segments that have not yet been executed are replaced with newly generated paths, updating the trajectory instruction set. This limits the correction to the area that causes the offset, without affecting stable segments that have been completed or are being executed. When replacing segments, 2 to 3 control points are reserved at the beginning and end of the replacement segment to overlap with adjacent segments, so that the laser execution unit transitions along the same path, avoiding step changes in the trajectory. For segments that are being executed, if an offset is detected but has not yet reached the threshold, only the offset trend is recorded, and the segment is not replaced immediately. It is then applied all at once when the segment ends, reducing execution jitter caused by frequent insertions. In areas where the number of online micro-correction triggers increases continuously in a short period of time, the control and communication unit can instruct the adsorption fixing unit to temporarily increase the adsorption intensity of the nearby partitions and decrease the intensity of the distant partitions, so that the edge of the shell is pressed back onto the bearing surface. This reduces the probability of further deviation from a mechanical perspective, thus decreasing the frequency of subsequent corrections.

[0056] A dual-scale strategy is introduced based on the cross-neighborhood: first, a short cross with a radius of 3 is used for rapid landing point prediction, and then a long cross with a radius of 7 is used for verification near the predicted point. Dual-scale can improve stability in noisy regions while controlling computational load. For the 5 pixels on both sides of the normal, not only the largest jump point is taken, but also the top 2 jump points are taken and a linear interpolation is performed between them to obtain sub-pixel positions closer to the true boundary; the interpolated results are still stored in pixel form, which facilitates consistent processing with other discrete points. In addition to the dual-sided grayscale jump and skeleton distance, consistency with the local ring code string can be added as a third reference: if the sector coverage of a certain path is empty in the local ring code string for a long time, then the segment is preferentially judged as a phantom segment. This can further reduce false detections on highly reflective materials. Multiple candidate entry points are explored along each path with a step size of 10 degrees. The product of the texture contrast and adsorption channel density on both sides of the normal of each candidate position in the empty reference image is calculated, and the position with the larger product is selected as the entry point. This approach balances both structural stability and visual recognition stability.

[0057] The effective processing area of ​​the base platform's bearing surface is set to 120×60. The adsorption and fixing unit is divided into 8×4 partitions with connected adsorption channels. The resolution of the visual acquisition unit is 1920×1440. The pixel coordinates of the fixed marker points in the image coordinates are P1(320,240), P2(1720,235), P3(1710,1230), and P4(330,1240). The four corner points in the bearing surface coordinates are Q1(0,0), Q2(120,0), Q3(120,60), and Q4(0,60). The laser execution unit has a fixed emission direction and ensures that the laser focus falls on the bearing surface. The control and communication unit is electrically connected to each unit through a single wire bundle. First, the adsorption and fixing unit is started in the order of "outer edge first, then middle" to form a stable bond and acquire the no-load reference image and the working image. Homography is established in the control and communication unit: let the image pixel coordinates be x = [uv1]. T (Where u represents column pixels and v represents row pixels), and the coordinates of the bearing surface are X = [xy1]. T (where x and y are the rectangular coordinates of the bearing surface), the homography matrix is... (Mapping the image onto the carrier surface), with a homogeneous scale of λ (used to eliminate homogeneous scaling), satisfying λX = Hx, and so on. Seeking Based on this, the working image is adjusted to the coordinates of the bearing surface, and background subtraction, connected component extraction, and hole filling are performed to obtain a binary mask containing only the main body of the shell and a set of pixels representing the shell boundary.

[0058] Next, the binary mask is eroded sequentially until the first minimal pixel cluster appears, resulting in the final set of retained pixels. (in (Including all mask pixels that were not deleted in the last round), the mask center point is defined as (u c ,v c ),calculate (in (where u is the cardinality of the set), this example yields (u) c ,v c = (1018, 738); Generate concentric rings with the center point of the mask as the center and a unit pixel radius until the outer shell boundary is covered. Let the pixel radius from any boundary pixel (u, v) to the center be... (where ρ is the radius pixel distance), maximum radius ρ max =512, therefore the number of concentric rings is 512; divide each concentric ring into 120 sectors, the angular width of a single sector Δθ = 360 / 120 = 3 degrees, let the direction angle θ(u,v) = atan2(vv c ,uu c (where θ is the angle, and atan2 is calculated in degrees), sector index

[0059] Generate a ring code string s for each concentric ring of radius r. r =[s r,0 ,…,s r,119 (where s) r,j ∈{0,1} indicates whether the j-th sector is occupied or not. When any boundary pixel satisfies round(ρ(u,v))=r and k(u,v=j, set s. r,j =1, and for the case of "both sides of a single empty sector are occupied", a bridging setting of 1 is performed to repair the gap. In the example, when r=508, the occupied segment interval is approximately [2,15], [28,43], [57,69], [82,95],

[0060] [101,118] etc.; then, the number of occupied segments and the length of each segment were counted for all ring code strings, and the start and end sectors were recorded. The ring with the largest total occupied length was selected as the outer boundary representative ring code string r. out =508, the inner ring, representing the ring code string r, occupies the smallest but non-zero total length. in =41, the ring occupying the middle position of the total length is the middle ring, representing the ring code string r. mid =276.

[0061] Constructing topological fingerprint digests (in This is a set of start and end sectors for each occupied segment. (Set of lengths for each occupied segment); refine the binary mask to obtain the skeleton pixel map, select the boundary pixel with the largest horizontal coordinate on the outer boundary of the shell as the skeleton in point, use eight-neighbor traversal to generate the skeleton walking trajectory and construct the skeleton symbol sequence Σ=[σ1,…,σ M ](in

[0062] σ i ∈{straight, left turn, right turn, fork in, fork out, terminate} represents a step event. After coarse screening is completed with Φ in the shell template library, the template skeleton symbol sequence Σ of the candidate templates is... (j) Aligning a template with only "match, merge, skip" operations with Σ, let the number of atomic operations for a given template be C. j (where C) j (This is a count of the sum of matches, merges, and skips), with M matching times. j (where M) j (The number of successfully matched symbolic pairs) is represented by the score vector S. j =(-C j M j ) Selecting the optimal template for lexicographical order

[0063] in accordance with The template key point list is a collection of template key points {q i}(where p) i (key points in the template coordinate system) and the set of key points in the current image {q} i}(where q) i To establish a one-to-one correspondence between key points in the image coordinate system, and to perform piecewise linear deformation by constructing a triangular mesh for every three adjacent key points: for any template path point p, in triangle △(p a ,p b ,p c Write the coordinates of the barycenter (α, β, γ) (where α + β + γ = 1 and α, β, γ ≥ 0) in the inner square and map them to the current image path point q = αq. a +βq b +γq c (where q) a ,q b ,q c For the corresponding key points in the current image, the same interpolation is used on the shared edges to ensure continuity. There are 40 key points on the outer boundary of the instance, 20 key points on each of the two holes, and a total of about 74 triangles. The path sketch generated by the mapping is discretized with a step size of 2 pixels and the boundary is fine-tuned. Let the tangential unit vector at a certain discrete point be t (where t is the unit vector along the path forward direction) and the normal unit vector be n (where n is the unit vector perpendicular to t). Take 5 pixels on the inside and 5 pixels on the outside along the n direction to form a cross neighborhood and calculate the gray level jump amplitude on both sides. (in and (The gray levels of the k-th pixels are the inner and outer edges, respectively), select max(Δ) in ,Δ outThe corresponding pixels are used as local boundary points, and the path row and column indices are smoothed by arithmetic average using a sliding window of length 7.

[0064] The smoothed path is verified point by point for the existence of gray-level jumps on both sides, the nearest distance d to the skeleton (where d is the minimum pixel distance from the path point to the skeleton pixel) satisfies d≤3, and it does not intersect with the verified path. If the violation occurs, the entire segment is marked as a phantom segment and deleted. If a break occurs, 3 to 5 sampling steps are taken along the tangent at both ends of the break to generate candidate bridging segments and the three conditions are repeated for screening until the path is connected and there are no new phantom segments. The continuous paths of the outer boundary and each hole position are sorted as "hole position first, then outer boundary". The adjacent holes with close distances are merged into the same group for continuous processing. An entry point and an exit point are set for each path so that the two points differ in angle by at least 20 degrees. Then, the homography mapping H is used to transform the path from image coordinates to bearing surface coordinates to form a cutting path set. A control point spacing of 1.0 is used in long straight sections, and a control point spacing of 0.4 is used in small radius corners and hole edges. Each path is divided into several segments (each segment has 10 to 18 control points) and organized in sequence into a trajectory instruction set.

[0065] During execution, the control and communication unit sends out trajectory instruction sets segment by segment. The vision acquisition unit acquires a local image with a radius of 50 pixels centered on the current cutting position every time a segment is completed or every 50 milliseconds. The concentric loops and sectors are quickly recoded to obtain the local loop code string s. loc (where s) loc The occupancy string (which only covers a local area) is compared sector by sector with the corresponding local ring code string of the template to count the number of mismatched sectors E (where E is the count of inconsistent sectors in the current window). When E>5, boundary fine-tuning, continuous path verification and phantom segment processing are re-executed only in this local area, and the corresponding segments that have not yet been executed are replaced. The first and last segments of the replaced segments overlap with the adjacent segments by 3 control points to ensure a continuous transition. If the same area is triggered for replacement twice in a row, the adsorption intensity of the adjacent partitions in this area is temporarily increased and the intensity of the far partition is reduced to suppress the offset again. In the example, an outer boundary path has about 12 segments and a total of about 860 control points, and two holes each have about 4 segments and a total of about 160 control points.

[0066] The present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. An adjustable laser cutting stage for mobile phone casings, characterized in that, include: The system comprises a base platform, an adsorption and fixing unit, a vision acquisition unit, a laser actuation unit, and a control and communication unit. The base platform is a planar load-bearing structure with a continuous adsorption channel. The adsorption and fixing unit is connected to the base platform to form partitioned adsorption and fixing on the load-bearing surface. The vision acquisition unit is perpendicular to the load-bearing surface of the base platform with a fixed line of sight, covering the entire area of ​​the load-bearing surface. The laser actuation unit is offset relative to the vision acquisition unit with a fixed emission direction, and its laser focus falls on the load-bearing surface of the base platform. The control and communication unit is fixed inside the base platform and communicates via a single... A wire bundle is electrically connected to the adsorption and fixing unit, the vision acquisition unit, and the laser execution unit to exchange shell shape data, cutting path set, and status data. The control and communication unit has a built-in shell template library and path generator. After receiving the original image from the vision acquisition unit, it sequentially performs image preprocessing and contour extraction, anchor block retrieval and matching based on the shell template library, calls the path generator to generate a cutting path set, converts the generated cutting path set into trajectory instructions suitable for the laser execution unit and issues them, and performs closed-loop control with online micro-correction based on the status data returned from the vision acquisition unit.

2. The adjustable laser cutting stage for mobile phone casings as described in claim 1, characterized in that, The phone case is placed flat on the support surface of the base platform and stably attached by the adsorption fixing unit; the vision acquisition unit first acquires an empty reference image, and then acquires a working image after the phone case is placed; the control and communication unit performs projection geometry tuning on the working image based on the fixed marker points on the support surface to obtain the homography mapping from the working image to the coordinates of the support surface; using the acquired empty reference image as the background, the method of background subtraction, connected component extraction and hole filling is used to obtain a binary mask containing only the shell entity and the set of shell boundary pixels.

3. The adjustable laser cutting stage for mobile phone casings as described in claim 2, characterized in that, The control and communication unit performs successive erosion on the obtained binary mask until a single pixel or a cluster of extremely small pixels appears for the first time. It records the arithmetic mean of the row and column indices of the formed pixel set and rounds it to the nearest pixel as the center point of the mask. Using a defined mask center point as the center, concentric rings are generated with increasing unit pixel radius until the outer shell boundary is covered. Each concentric ring is divided into 120 sectors, with sector numbers starting from a fixed reference direction in a clockwise direction. For each sector of each concentric ring, the presence of an outer shell boundary pixel is checked. If an outer shell boundary pixel is present, it is marked as occupied; otherwise, it is marked as empty. The 120 occupied and empty states of a concentric ring are connected in sector number order to form a ring code string of fixed length. The above operation is performed on all concentric rings to obtain a set of ring code string sequences.

4. The adjustable laser cutting stage for mobile phone casings as described in claim 3, characterized in that, The control and communication unit counts the number of occupied segments and the continuous length of each occupied segment in each ring code string, and records the start and end positions of each occupied segment according to the sector number; it selects three types of representative ring code strings from all generated ring code string sequences: the outer ring code string with the largest total occupied segment length, the inner ring code string with the smallest but non-zero total occupied segment length, and the middle ring code string with the middle total occupied segment length; it then connects the occupied segment position list and length list of the three selected representative ring code strings in a predetermined order to form a topological fingerprint digest.

5. The adjustable laser cutting stage for mobile phone casings as described in claim 4, characterized in that, The control and communication unit refines the binary mask of the current mobile phone shell to obtain the skeleton pixel map; the boundary pixel with the largest horizontal coordinate after projection adjustment is selected as the skeleton in point on the outer boundary of the shell. The skeleton walking trajectory is generated by traversing eight neighborhoods. During the skeleton traversal, the relative position of the current pixel and the previous pixel is used as the basis for progressive encoding: if the position of the next pixel remains unchanged from the previous one, it is recorded as a straight-line symbol; if the next pixel is in the left neighborhood, it is recorded as a left turn symbol; if it is in the right neighborhood, it is recorded as a right turn symbol; when a fork is encountered, a branch stack is pushed sequentially according to the neighborhood order in which the fork appears, and a fork entry symbol is recorded when entering a branch and a fork exit symbol is recorded when returning to the main trunk; a termination symbol is recorded when reaching the endpoint; after completing the traversal, a skeleton symbol sequence covering the outer boundary and internal hole positions is obtained.

6. The adjustable laser cutting stage for mobile phone casings as described in claim 5, characterized in that, For each template in a candidate template set, the control and communication unit stores offline a sequence of skeleton symbols representing its template outline and a list of key points for the template. The list of key points includes the symbol number of the outer boundary start point, each bifurcation position, and each hole entrance position. Sequence alignment is performed between the generated current skeleton symbol sequence and the skeleton symbol sequence of each template. Allowed atomic operations include only matching, skipping, and merging adjacent symbols into a single transition symbol. During alignment, matching is preferred, followed by merging, and then skipping. When multiple feasible alignment paths exist, the one with the fewest total atomic operations and the most matching times is selected as the alignment result of the corresponding template. After aligning all candidate templates, the template with the best score is selected as the final template. Based on the key point list of the final template and the corresponding symbol number in the alignment result, a one-to-one correspondence is established between the key points of the template and the key points of the current image. Piecewise linear deformation is performed inside the triangular mesh formed by every three adjacent key points to map the template path sketch to the coordinates of the current image. Interpolation is used between the triangular meshes on the shared edges to ensure the continuity of the path.

7. The adjustable laser cutting stage for mobile phone casings as described in claim 6, characterized in that, The control and communication unit performs boundary fine-tuning, specifically including: constructing a cross-shaped neighborhood perpendicular to the tangent of the path along each discrete point of the mapped path sketch; taking several pixels inward and outward from the constructed neighborhood, comparing the gray-level jump amplitude of the pixels on both sides, and selecting the pixel with the largest jump as the local boundary landing point; connecting all local boundary landing points into a continuous path in the original order; performing an arithmetic mean of the row and column indices of the path points at the corner using a sliding window of fixed length to obtain a smooth corner; projecting the generated continuous path back to the working image, and checking whether each segment on the path simultaneously satisfies the following conditions: gray-level jump exists on both sides, the distance to the nearest skeleton segment is less than the preset pixel distance, and it does not intersect with other verified paths; path segments that do not meet any of these conditions are marked as phantom segments and deleted; if the deletion operation causes the path to break, then bridging segments are repeatedly generated in the neighborhood of the broken endpoint until the path is connected again.

8. The adjustable laser cutting stage for mobile phone casings as described in claim 7, characterized in that, The control and communication unit sorts the continuous paths that have passed the phantom segment test according to the outer boundary and each hole position, and specifies the cutting in and cutting out order to form the final cutting path set; using the obtained homography mapping, the final cutting path set is transformed from image coordinates to bearing surface coordinates and stored as a trajectory instruction set.

9. The adjustable laser cutting stage for mobile phone casings as described in claim 8, characterized in that, The control and communication unit sends the generated trajectory instruction set segment by segment to the laser execution unit for cutting; the vision acquisition unit acquires local images at fixed intervals, and the control and communication unit performs rapid recoding of concentric rings and sectors in a small area around the current cutting position, comparing it with the local ring code string of the corresponding template; when the number of mismatched sectors in a local area exceeds a preset threshold, boundary fine-tuning is re-executed only in that local area and the trajectory instruction set to be executed is updated; the adsorption and fixing unit maintains stable adhesion throughout the entire cutting process until all trajectory instruction sets are executed.

Images