Vision algorithm and device adaptation·Capability overview
Vision Algorithm Modules and SDK Development Services
Algorithm development for equipment manufacturers, system integrators and enterprise R&D teams, covering image processing and measurement, laser spot and laser line analysis, and infrared and thermal imaging. The scope of development, SDK delivery, device integration and maintenance is defined from the samples, hardware, interfaces and acceptance requirements.
Select algorithm services by task
Image Processing, Measurement and Laboratory AnalysisDroplet contact angle, 2D dimensions, contours, registration, segmentation, timing and measurement evaluation.Laser Spot and Laser Line AlgorithmsLight spot positioning, light beam profile analysis, calibration and potential application scene directions are evaluated according to actual imaging conditions.Infrared and Thermal Imaging AlgorithmsData correction, heat map analysis, object detection, multi-modal registration and industry application fields.
Service scope
Vision algorithm projects usually involve imaging, data, algorithms, interfaces, and device operating environments simultaneously. The service starts from the customer's detection or measurement task and combines the required capabilities into an executable solution.
| Customer needs | Service content | main output |
|---|---|---|
| New device algorithm development | Sample evaluation, solution design, algorithm combination, interface development and equipment joint debugging | Project software, parameter configuration, interface documentation and verification report |
| Optimization of existing equipment | Problem recurrence, sample analysis, parameter or algorithm adjustment, regression of old and new versions | Optimized version, problem records, comparison results and upgrade instructions |
| Algorithm module integration | Input and output confirmation, library or program access, exception status and result format adaptation | Calling interface, examples, status codes and deployment instructions |
| Multi-device replication deployment | Fixed software and configuration, device licensing, installation checks and version management | Licensing records, deployment packages, device ledgers and recovery plans |
| Continuous maintenance and iteration | Original scenario problem handling, regression testing, controlled updates and change evaluation | Maintain records, release notes and iteration plans |
Key algorithmic capabilities
Algorithmic capabilities are organized by enterprise mission. A project can combine multiple modules and add project adaptations based on imaging, calibration, business rules and operating platforms.
| Ability page | Applicable tasks | Typical output | Project confirmation key points |
|---|---|---|---|
| Laser Spot Localization and Beam Profile Analysis | Single point, multi-point and light spot detection | Center coordinates, morphological parameters, quality status and calibration results | Wavelength, exposure, spot size, reference position and equipment platform |
| Contact Angle and Droplet Profile Analysis | Droplet profile, baseline and dynamic process analysis | Contact angle, fitting parameters, geometric quantities and contact line status | Droplet models, baselines, scales, test conditions and reference methods |
| 2D Dimensional and Contour Measurement | Edge, contour and 2D point set measurements | Position, size, fit residuals, and shape deviations | Drawings, tolerances, datums, calibrations, optical conditions and reference quantities |
| Image Registration and Camera Calibration | Image alignment, 2D mapping, calibration and dedistortion | Transformation matrix, camera parameters, distortion parameters and residuals | Coordinate system, transformation model, calibration plate, field of view and recalibration conditions |
| Image Segmentation and Particle Analysis | Region extraction, particle counting, adhesion separation and morphological statistics | Masks, objects, counts, outlines, skeletons and distributions | Foreground definition, annotation rules, glue, boundary objects and scales |
| Motion, Time-Series and Process Analysis | Movement estimates, trends, cycles, events and anomaly scoring | Displacements, trajectories, spectra, periods, events and scores | Timestamp, sampling rate, scale, operating conditions, normal baseline and event definition |
| Wetting Properties and Interfacial Parameter Estimation | Surface energy, critical tension, powder imbibition and hanging drop calculations | Physical property components, regression results, residuals and status | Models, liquid properties, units, temperature, calibration and reference methods |
| Measurement Repeatability and Uncertainty Assessment | Measurement system analysis, calibration regression and uncertainty calculation | Variance components, intervals, regression parameters, and uncertainty | Design of experiments, reference equipment, measuring ranges, tolerances and error models |
| Equipment authorization and continuous optimization services | First prototype development, equipment duplication, maintenance and version upgrades | License scope, version ledger, maintenance records and change plans | Number of devices, scenarios, deadlines, replacement, offline operation and service scope |
Project selection method
- Define business results: Determine whether coordinates, dimensions, angles, counts, masks, trends, physical parameters, or pass judgment are required, and describe the units and downstream uses.
- Provide representative data: Submit normal, borderline, and abnormal samples, along with field of view, resolution, exposure, working distance, sampling rate, and field variations.
- Determine the evaluation method: Clarify reference values, labeling rules, sample stratification, statistical denominators and passing thresholds.
- Confirm the operating environment: Describe the operating system, CPU/GPU/NPU, memory, calling language, interface, concurrency, number of devices and offline requirements.
- Complete sample verification: Check the effect, failure handling, time consumption and resource usage on independent data, and then determine the scope of development and deployment.
Existing equipment upgrade projects also need to provide the current software version, problem samples, operation logs, camera and light source configurations, and recent hardware change records.
Sample verification and performance confirmation
Algorithm performance is directly related to imaging conditions, sample distribution, target platform and evaluation caliber. The project adopts phased verification, so that the accuracy, detection rate, false alarm rate, processing rhythm and stability have a clear scope of application.
| stage | main work | Project deliverables |
|---|---|---|
| Data check | Check original data, reference values, abnormal samples and evaluation rules | Enter checklist, gaps and verification plan |
| sample baseline | Compare algorithm routes and record parameters, results, and failure conditions | Feasibility conclusions, baseline results and implementation recommendations |
| project development | Complete module combination, interface, configuration, status and log | Runnable version, interface information and test records |
| Independent data acceptance | Use data that has not participated in parameter adjustment or retest at an agreed work station | Indicator statistics, sample-by-sample results and acceptance records |
| Equipment operation verification | Check continuous operation, exception recovery, resource and version management | Run reports, deployment instructions, and fallback scenarios |
Performance metrics are defined at project initiation and documented along with sample range, hardware configuration, software version, and test conditions.
112 Algorithm Capability Catalog
The table below is used for preliminary technology selection and lists the main inputs, available outputs and project confirmation points. Final interface fields, units, thresholds, platforms, and acceptance criteria are determined in the project plan.
F01 Droplet profile and contact angle (9 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A01 | Height-width method (θ/2) | Droplet contour, contact baseline and coordinate convention | Contact angle and geometric quantities used for review | Approximation of spherical cap applies; not arbitrary drop shape model |
| ALG-A02 | Algebraic Least Squares Circle Fitting | Seat drop outline and contact point definition | Fitting circle parameters and contact angle | Overall arc assumption; need to check the influence of gravity deformation |
| ALG-A03 | Ellipse fitting | 2D contour points | Ellipse center, axis length and direction | Used to output geometric ellipse parameters; contact angle needs to be calculated based on baseline, contact point and angle definitions |
| ALG-A04 | Robust local polynomial fitting | Droplet contour, polynomial order and fitting interval | Contact point tangent and angle | The fitting interval and baseline must be confirmed based on the sample |
| ALG-A05 | Smooth cubic B-spline fit | Droplet contour, control points and smoothing parameters | Smooth contours, local tangents and angles | Smoothing cannot restore the true occluded contours |
| ALG-A06 | Young-Laplace / ADSA | The contour of a seat drop that satisfies the conditions of axial symmetry and static equilibrium | Zuodi model fitting parameters and angles | The seated drop model does not automatically become the hanging drop image inversion product |
| ALG-A28 | Numerical integration of axisymmetric contours | Axial and radial profiles, length scales | Axisymmetric volume and surface area | Requires axial symmetry and reliable contours; physical quantities depend on scale |
| ALG-A92 | Tilt baseline coordinate correction | Contour points, baseline endpoints and image axis directions | Correction coordinates and transformation | Coordinate correction cannot fix wrong contact points |
| ALG-A93 | Contact line motion status identification | Time, contact radius, angle and speed threshold | Contact line motion status and segmentation | Status determination is limited by sampling rate and threshold |
F02 Geometric Measurement and Profile Analysis (16 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A07 | Total least squares straight line fit | Two-dimensional point set in the same coordinate system | Line direction, position and fitting residuals | Least squares results do not automatically exclude outliers |
| ALG-A08 | RANSAC Robust model estimation | 2D point set, model type and interior point threshold | Line or circle models, interior points and residuals | The results can only be compared after the model type and threshold are clear. |
| ALG-A09 | Hough linear detection | Grayscale or edge map, length and connection thresholds | Candidate line segments and locations | Candidate line segment is not an identified product edge category |
| ALG-A10 | Sub-pixel gradient caliper | Image, caliper sampling line and edge polarity | Sub-pixel edge position on sample line | Pixel position does not directly equal physical size accuracy |
| ALG-A17 | Convex Hull Rotating Caliper Method | Contour points in the same coordinate system | Maximum and minimum Feret diameter | Physical dimensions require scaling |
| ALG-A18 | Normal section double edge width measurement | Binary mask, centerline and normal sampling parameters | Section width along centerline | 2D width cannot be interpreted as trench depth |
| ALG-A19 | Analysis of Symbol Distance Field Contour Deviation | Registered reference and observation masks | Signed distance and contour deviation | Registration error must be distinguished from shape error |
| ALG-A29 | Geometric orthogonal distance circle fitting | Two-dimensional near-circle point set and iteration parameters | Geometric distance fitting circle and residual error | The convergence result is affected by the initial value, point set quality and outlier proportion. |
| ALG-A30 | Huber IRLS robust straight line fitting | Two-dimensional point set, iteration and Huber adjustment parameters | Robust straight line and weight information | Robustness has scope of application and cannot replace sample quality inspection |
| ALG-A34 | Chamfer distance template matching | Image and template edges | Chamfer matching cost and candidate positions | Matching distance is not class probability |
| ALG-A35 | Hu moment invariant contour matching | Two 2D contours and how to compare them | Hu shape difference score | Score is not equivalent to calibrated identification confidence |
| ALG-A36 | Fourier contour descriptor | Ordered closed contours, number of samples and number of harmonics | Fourier contour descriptor | The descriptor does not automatically provide product classification labels |
| ALG-A41 | Three-dimensional point set orthogonal plane fitting | A three-dimensional point set with physical meaning | Plane normal, position and residual | Three-dimensional points cannot be generated out of thin air from ordinary two-dimensional photos. |
| ALG-A45 | Bidirectional Hausdorff and Chamfer contour distance | Two contours in the same coordinate system | Two-way Hausdorff and Chamfer distance | Distance thresholds must be defined based on product tolerances and sampling |
| ALG-A46 | Minimum Area Rotated Envelope Rectangle | 2D point set | Rotate envelope rectangle, orientation and size | The outer rectangle does not replace all dimensional and geometric tolerances |
| ALG-A99 | Two-dimensional minimum area straightness | Limited two-dimensional sampling points | Minimum envelope parallel band width and direction | Does not directly constitute a complete set of ISO or ASME geometric tolerance inspections |
F03 Particle segmentation and image enhancement (32 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A11 | Otsu Otsu Threshold Segmentation | Grayscale, inverse color and smoothing settings | Global threshold, binary mask | Grayscale distribution that needs to be suitable for global threshold |
| ALG-A12 | Connected domain labeling and filtering | Binary mask, area lower limit and connectivity rules | Area list, size and location | Connected domains cannot automatically distinguish product categories |
| ALG-A13 | Marking control watersheds | Image, foreground mask or seed parameters | Watershed areas and labels | Seed and foreground quality determine detachability conditions |
| ALG-A14 | Normalized cross-correlation template matching | Reference templates and images to be searched | Match scores and candidate locations | Rotation, scaling or perspective changes need to select the corresponding registration method and verify it separately |
| ALG-A15 | Zhang-Suen skeleton refinement | Binary mask and iteration upper limit | single pixel skeleton | Skeletons do not automatically include defect semantics |
| ALG-A16 | Euclidean distance transform | binary mask | Pixel distance field | 2D mask distance is not equal to physical depth |
| ALG-A27 | U-Net semantic segmentation | Legally available product domain ONNX weight and preprocessing definition | Segmentation mask and interface agreement results | Projects need to provide or train model weights for applicable product categories, and confirm weight licensing and product domain verification |
| ALG-A43 | Multi-scale LoG blob detection | Image, multi-scale and response thresholds | Spot candidate center, scale and response | Candidate scale is not directly equal to physical particle size |
| ALG-A44 | DBSCAN density clustering | Point or feature, neighborhood radius and density lower limit | Cluster labels and noise points | Clustering cannot automatically name defect types |
| ALG-A47 | Adaptive Gaussian Threshold | Grayscale, windows and offsets | Local threshold and binary mask | The parameters need to be combined with the defect scale and cannot be applied to any texture. |
| ALG-A48 | Sauvola local threshold | Grayscale image, window and local statistical parameters | Local threshold map and mask | Requires independent marking and confirmation, not unconditional defect detection |
| ALG-A49 | Triangular threshold segmentation | Grayscale image and inverse color settings | Triangulation Thresholding and Masking | Imaging distribution that needs to comply with the threshold model |
| ALG-A50 | GrabCut foreground segmentation | Color image and rectangle or initial label | Foreground labels and masks | Needs initialization, not product domain semantic model |
| ALG-A51 | K-means color segmentation | Color map, color space and number of clusters | Color cluster labels and related results | Color clusters are not equivalent to pass and defect categories |
| ALG-A52 | Chan-Vese level set | Grayscale image, initial mask and evolution parameters | Region masks and iteration results | Complex textures need to be verified using other methods or combinations |
| ALG-A53 | LoG zero crossing edge | Grayscale image, filter scale and contrast threshold | zero crossing edge | Edge response does not automatically form dimensional acceptance results |
| ALG-A54 | Scharr gradient edge | Grayscale image and gradient threshold | Gradient response and edges | Gradient intensity does not directly correspond to crack depth |
| ALG-A55 | Multi-scale top-hat background correction | Images, multi-scale structural elements and target light and dark properties | Background correction and multi-scale response | Processing does not demonstrate the physical properties of the target material |
| ALG-A56 | CLAHE local contrast enhancement | Images, Contrast Limits and Grids | Enhance images | The enhanced appearance changes are not new measurement information |
| ALG-A57 | Single-Scale Retinex Illumination Correction | Image and lighting scale | Retinex correction chart and related results | Does not recover information lost due to occlusion or saturation |
| ALG-A58 | Bilateral filtering | Image, neighborhood and bilateral weights | Denoised image | The denoising effect must be evaluated in conjunction with the measurement chain |
| ALG-A59 | Non-local mean denoising | Image, noise intensity and search window | Denoised image | No commitment to recover structures below imaging resolution |
| ALG-A60 | Morphological reconstruction | Seed graph, constraint graph and connectivity rules | Reconstruction results and iteration status | Seed errors will change the results and cannot replace semantic judgment. |
| ALG-A61 | Binary hole filling | Binary mask and connectivity rules | Hole filling mask and processing results | If real holes are detection targets, they cannot be erased by preprocessing. |
| ALG-A62 | Distance field h-maxima seeds | Binary mask, h value and minimum interval | Distance field seeds and related results | Only segmentation seeds are generated, which does not mean that the targets have been correctly separated. |
| ALG-A63 | Skeleton graph path tracing | Single pixel skeleton or explicit refinement options | Skeleton nodes, branches and paths | Path tracing does not automatically differentiate between cracks, scratches and dust |
| ALG-A64 | Dynamic planning of boundary paths | 2D cost map, path axis and endpoint constraints | Constraint paths and costs | Path dependence cost map, not universal contour ground truth |
| ALG-A65 | Grayscale center of gravity laser center line | Laser line image, scanning axis and background parameters | Grayscale center line | The pixel centerline is not equal to the calibrated 3D contour |
| ALG-A66 | Gabor texture features | Image, direction and wavelength filter parameters | Gabor texture response and characteristics | Feature extraction does not automatically complete defect classification |
| ALG-A80 | Local binary pattern LBP texture | Grayscale image | LBP coding and texture statistics | Coding itself has no defect category meaning |
| ALG-A81 | GLCM gray level co-occurrence matrix texture | Quantized grayscale images, levels and spatial offsets | Co-occurrence matrix and texture features | Feature thresholds need to be verified according to tasks |
| ALG-A82 | Frequency domain Gaussian notch to remove periodic fringes | Image, fringe frequency position and notch bandwidth | Deperiodic fringe image | Need to distinguish true repeating structures from interference fringes |
F04 Registration camera and laser geometry (10 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A20 | Steger sub-pixel ridge extraction | Grayscale linear image, scale and curvature parameters | Sub-pixel ridge position and associated response | Ridge extraction is not a finished 3D scanning system |
| ALG-A21 | Laser triangulation | Camera internal and external parameters, laser plane and image position | The three-dimensional intersection point of the ray and the calibration plane | Must be calibrated using actual equipment, physical accuracy cannot be delivered with synthetic parameters |
| ALG-A31 | Phase-dependent translation registration | Reference and mobile images with overlapping textures | 2D translation and related responses | Suitable for translation estimation; rotation or three-dimensional posture requires selecting the corresponding model |
| ALG-A32 | ECC intensity registration | Overlay images, motion models and initial values | Specify two-dimensional transformations and related indicators | Relying on the initial value and image similarity, the project needs to set a failure determination |
| ALG-A33 | ORB and RANSAC homography registration | Source and target images with sufficient features | Feature correlation and homography transformation | Applicable to homography; does not automatically cover multi-depth scenes |
| ALG-A37 | 2D Rigid Body ICP Registration | 2D point set, approximate initial pose and pruning parameters | Two-dimensional rigid body transformation and registration residuals | It is not a global registration, nor is it a 3D robot hand-eye calibration. |
| ALG-A38 | Corresponding points should be fitted | Corresponding points on the same plane | Two-dimensional homography matrix | Single plane relationships cannot be substituted for complete three-dimensional calibration |
| ALG-A39 | Multi-view checkerboard camera calibration | Multiview checkerboard or corner points, dimensions and image specifications | Camera internal parameters, distortion and calibration results | The installation and lens status must be reviewed after changes. |
| ALG-A40 | Image and point dedistortion | Image or pixels, camera intrinsic parameters and distortion | Correct image or point coordinates | Relies on effective internal parameters and does not promise to eliminate all optical errors |
| ALG-A42 | SVD Rigid Body and Similarity Registration | Sequentially corresponding two-dimensional point set and scale options | Rigid body or similar transformation | Suitable for two-dimensional corresponding points; three-dimensional registration requires the establishment of independent data and coordinate links |
F05 Movement timing and process statistics (20 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A22 | Kalman filter | Initial position, time step and noise settings | Two-dimensional position and velocity state estimation | Filtered output cannot disguise invalid measurements as measured positions |
| ALG-A23 | hungary match | Cost matrix and rejection threshold | One-to-one associations and unmatched items | The association results do not constitute real identification. |
| ALG-A24 | Savitzky-Golay smoothing and derivation | Sequence, window, order, sampling interval and derivative order | Smooth or derivative sequence | Uniform sampling and parameter legality must be clear |
| ALG-A25 | Welch power spectrum estimation | Signal, sample rate and segmentation windows | Frequency and power spectral density | Long-term equipment health cannot be inferred from a single segment of spectrum |
| ALG-A26 | Mahalanobis distance anomaly detection | Normal training features, regularization and test features | Distance score and confirmed anomaly judgment | Needs true normal distribution, not a universal defect identifier |
| ALG-A67 | Lucas-Kanade sparse optical flow | Two consecutive frames, initial corner points and tracking parameters | Sparse point motion and effective state | Optical flow tracking does not equal cross-mirror identity correlation |
| ALG-A68 | Farneback Dense Optical Flow | Two consecutive frames and optical flow parameters | Dense two-dimensional displacement field | Image displacement is not directly equal to three-dimensional velocity |
| ALG-A69 | MOG2 background separation | Continuous frame and background learning parameters | foreground mask | Prospects are not specific categories such as people, flames, etc. |
| ALG-A70 | Dual frame differential motion detection | Two frames, difference threshold and morphological parameters | motion differential mask | Unable to independently judge the category or danger level of the moving target |
| ALG-A71 | Exponential Moving Average EWMA | Numeric sequence, smoothing coefficient and initial value | EWMA smoothed series | Smoothing cannot eliminate system drift |
| ALG-A72 | Hampel median absolute deviation anomaly filtering | Numerical sequences, windows and MAD thresholds | Outlier markers and processing sequences | Replacement values cannot impersonate original valid measurements |
| ALG-A73 | Bilateral CUSUM change detection | Sequence, target mean, allowable deviation and threshold | Change statistics and alarm candidates | Alarm thresholds and equipment disposal will be agreed separately. |
| ALG-A74 | short time Fourier transform STFT | Signal, sample rate, window and step size | time spectrum | Aliasing information cannot be recovered when the sampling rate is insufficient |
| ALG-A75 | FFT autocorrelation period estimation | Sequence, sample rate and search delay range | Period or frequency estimates and related quantities | The existence of cycles does not directly prove the type of equipment failure. |
| ALG-A76 | Dynamic Time Warping DTW | Two-sequence and path window | DTW distance and alignment path | Similarity distance does not equal true event identity |
| ALG-A77 | Recursive Least Squares RLS | Feature vectors, observations, and recursion parameters | Online model coefficients and status | Model and data incentive conditions need to be defined |
| ALG-A78 | Welford online mean and covariance | Dimensionally consistent observation sequence | Online mean, covariance and count | Statistics do not automatically give process qualification judgments |
| ALG-A79 | PCA reconstruction residual anomaly score | Normal samples, feature normalization and principal component scores | reconstructed residual difference fraction | It is necessary to establish a normal baseline, not a general fault diagnosis |
| ALG-A83 | Short time RMS envelope | Signals, windows and steps | Short time RMS envelope | Physical units depend on input calibration |
| ALG-A84 | One-dimensional peak significance detection | One-dimensional sequence, saliency and minimum interval | Peak position and peak correlation quantity | Peaks cannot be automatically named as defect events |
F06 Wetting physical property inversion (7 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A85 | OWRK Solid surface energy inversion | Multi-probe liquid contact angle and component physical properties | OWRK solid surface energy component | Consistent physical property units and applicable surface models need to be used |
| ALG-A86 | Wu Harmonic mean surface energy inversion | Contact angle and probe liquid dispersion and polar components | Wu surface energy component | Do not directly infer physical properties from a single uncalibrated graph |
| ALG-A87 | vOCG Acid-base surface energy inversion | Contact angle, probe liquid LW and acid-base content | vOCG surface energy component | Probe fluid combinations and model assumptions must be clear |
| ALG-A88 | Zisman Critical surface tension extrapolation | Series contact angle and liquid surface tension | Critical surface tension extrapolation and fitting | The extrapolated value is not equivalent to the true value of the universal solid surface energy |
| ALG-A89 | Washburn Powder liquid absorption contact angle | Time, liquid absorption quality, material constants and physical properties of SI liquid | Powder liquid absorption contact angle and regression amount | Material constants are calibrated with the same filling conditions; different agreed constants cannot be mixed. |
| ALG-A90 | Hanging drop Young-Laplace forward integration | Bond parameters, integration arc length and step size | Dimensionless pendant drop forward profile | It is not necessary to find the complete ADSA of the unknown surface tension from the photo. |
| ALG-A91 | Hanging drop curvature regression tension estimation | Pre-smoothed semi-contours, density differences and gravity in meters | Curvature regression tension estimation | The height direction must be consistent; it is not a complete nonlinear ADSA fitting |
F07 Measurement Statistics and Uncertainty (6 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A94 | Bootstrap percentile confidence interval | Samples, statistical functions, confidence levels, and resampling settings | Percentile intervals and statistical results | Confidence level setting does not equal actual coverage proof for finite samples |
| ALG-A95 | Monte Carlo measurement uncertainty propagation | Measurement model, input means and covariances, sampling settings | Model output distribution and uncertainty statistics | Missing system items are not automatically completed by sampling |
| ALG-A96 | Crossed ANOVA Gage R&R | Balance data by part, operator, number of repetitions | Gage R&R and variance components | Do not automatically incorporate interaction terms or perform full significance tests |
| ALG-A97 | Weighted least squares linear calibration | Reference x and measurement y and standard uncertainty | Linear calibration coefficients and residuals | Regression residuals are not full equipment uncertainties |
| ALG-A98 | Deming Bivariate Error Regression | Two variable observations and error variance ratio | Deming regression coefficient | The error variance ratio needs to have a basis |
| ALG-A100 | First-order sensitivity uncertainty propagation | Measurement model, mean, covariance and difference step size | Sensitivity and first-order propagation results | Strong nonlinearity and missing system quantities need to be evaluated separately. |
F08 Laser beam profile analysis (12 items)
| No. | Algorithm module | Common inputs | Output available | Project confirmation points |
|---|---|---|---|---|
| ALG-A101 | Adaptive threshold laser spot grayscale centroid | Grayscale frame, ROI, background and candidate filtering parameters | Single bright spot center, area and energy related quantities | The physical coordinate output needs to be verified with the target camera, lens and calibration data. |
| ALG-A102 | Multi-laser spot connected domain candidate extraction | Grayscale frame, ROI and connected domain filtering parameters | Multiple highlight candidates and their characteristics | Contact or overlapping bright spots do not automatically separate |
| ALG-A103 | Logarithmic quadratic surface sub-pixel spot center | Approximate Gaussian spot, window and background parameters | Logarithmic quadratic center and fitting results | Strong saturation, non-Gaussian and window truncation may produce bias |
| ALG-A104 | Gradient radial symmetry light spot center | Near-radial symmetric spot and gradient screening parameters | Radial symmetry center and residual | The positioning effect needs to be verified according to the reflective target shape and imaging conditions. |
| ALG-A105 | Laser spot second moment and main axis diameter | Linear response to spot intensity, background and ROI | Center of mass, second-order moment primary and secondary axis diameters and directions | Only second-order moment calculation is implemented, not a complete ISO11146 measurement system |
| ALG-A106 | Laser spot surrounding energy diameter | Spot, center, energy ratio and scale | Specified ratio of surrounding energy radius and diameter | Grayscale integral does not automatically convert to Watts or Joules |
| ALG-A107 | Laser spot quality and usability diagnostics | Spot, exposure related information and quality limits | Diagnostic quantities and limit checks such as saturation, truncation, and axis ratio | Quality thresholds require equipment testing and are not automatic acceptance criteria. |
| ALG-A108 | One-to-one matching of multi-laser spot reference bits | Detection points, reference points and distance upper limit in the same coordinate system | One-to-one matches and missing items | It does not estimate external parameters and does not prove the identity of the target. |
| ALG-A109 | Two-dimensional pixel to physical coordinate polynomial calibration | Pixels in the same plane as reference coordinates and mapping order | Two-dimensional mapping coefficients and fitting results | Two-dimensional empirical mapping does not replace three-dimensional calibration |
| ALG-A110 | laser spot calibration mapping and covariance propagation | Pixel points, effective 2D calibration and pixel covariance | Mapping coordinates and propagation of given covariances | Calibration plates, temperature drift and model deviations are not automatically included |
| ALG-A111 | laser spot pixel to camera view angle conversion | Pixels, camera intrinsics and distortion | Camera coordinate system unit line of sight and angle | Missing distance information, 3D points cannot be recovered individually |
| ALG-A112 | Laser spot two-dimensional repeatability and coverage ellipse | Repeat the same conditions for 2D coordinates and coverage levels | Mean, covariance, radial RMS and covering ellipse | The two-dimensional normal covering assumption does not prove physical accuracy or long-term stability |
From assessment to delivery
- Need confirmation: Clarify tasks, data, outputs, indicators, platforms, interfaces, number of devices and planned time.
- Sample verification: Fixed samples and evaluation calibers form the algorithm baseline, applicable conditions and implementation scope.
- Development and joint debugging: Complete algorithm combination, data access, parameter configuration, abnormal status, result interface and log.
- Independent acceptance: Verify the effect, time consumption, stability and recovery mechanism based on agreed data or equipment stations.
- Authorization and maintenance: Determine the running license according to the device, scenario and version, and continuously handle the original scenario problems and controlled upgrades.
The delivery content may include algorithm libraries or programs, calling examples, interface and parameter descriptions, configuration or calibration files, test reports, version notes and recovery plans. Source code, project-specific models, third-party components, hardware and new platform adaptations are confirmed separately in the quotation.
FAQ
How can I tell if an existing algorithm will work with my device?
First provide original samples, mission objectives, reference results and equipment conditions. After completing data inspection and sample verification, reusability, development that needs to be supplemented, expected deliverables, and acceptance methods can be clarified.
Are Windows, Linux and embedded platforms supported?
Adaptation can be evaluated based on the target system, processor, and calling interface. The project will also confirm dependencies, resources, delivery form, installation method and performance test conditions.
Why do accuracy and processing speed need to be confirmed on a project-by-project basis?
Indicators will vary with sample difficulty, image resolution, algorithm combination, hardware and statistical caliber. Through customer samples and target platform testing, acceptable project indicators can be formed.
Is it possible to license based on the number of devices?
Licenses can be agreed based on customer entities, equipment or workstations, algorithm scenarios, delivery versions, and deadlines. The rules for machine replacement, offline operation, recovery and new equipment are specified in the contract.
Can we continue to optimize in the future?
It can reproduce, improve and regression verify new samples, misjudgment problems and changes in working conditions in the original scene. New products, hardware or features are evaluated based on the scope of change.
Project consultation
xuzhiyang0928@gmail.com
Define the Development Scope with Samples and Device Requirements
Vision Algorithm Modules and SDK Development Services
Please describe the task object, original picture or video, target indicators, device platform, interface and sample situation.
Phone: 13910119357xuzhiyang0928@gmail.com