Machine Vision Inspection Patents

Machine vision inspection patents cover 2D pattern matching geometric and grayscale template alignment innovations; 3D structured light laser triangulation and time-of-flight depth sensing innovations; deep learning convolutional neural network defect classification and segmentation innovations; and line scan TDI CCD high-speed area scan camera innovations — with IP held by vision system OEMs, camera manufacturers, and semiconductor inspection equipment companies: MAJOR MACHINE VISION PATENT HOLDERS: COGNEX: 2,000+; specific inspection innovations (specific specific PatMax geometric pattern matching: specific specific trained geometric model from specific specific 1 μm accuracy 0.01° rotation from specific specific illumination invariant from specific specific partial occlusion 80% from specific specific VisionPro SDK: specific specific CogFindCircleTool from specific specific CogBlobTool from specific specific CogCalibCheckerboardTool from specific specific In-Sight 9000: specific specific 64-bit Intel i7 from specific specific 12-megapixel Sony IMX from specific specific GigE Vision from specific specific < 30 ms cycle time from specific specific ViDi deep learning: specific specific ViDi-Classify ResNet backbone from specific specific ViDi-Detect YOLO-v5 from specific specific ViDi-Segment U-Net from specific specific 100-500 images training from specific specific 99.99% uptime from specific specific SMT PCB IC package BGA from specific specific 2D barcode OCV verify read from specific specific Cognex Mobile Barcode 1D 2D QR Data Matrix); KEYENCE: 1,000+; specific sensor innovations (specific specific CV-X series 2D: specific specific 16-megapixel imager from specific specific 320 fps from specific specific color monochrome from specific specific IX-G: specific specific 4-axis sync from specific specific 0.1 pixel accuracy from specific specific 3D-A7000: specific specific chromatic confocal from specific specific 0.02 μm Z accuracy from specific specific LJ-X8900 laser profile: specific specific 408 nm blue semiconductor laser from specific specific 3.2M pts/s from specific specific 0.1 μm Z from specific specific XG-X2800 3D: specific specific pattern projection stereo from specific specific 0.5 mm FOV from specific specific 100 ms 3D capture); TELEDYNE DALSA: 500+; specific camera innovations (specific specific Falcon4 25MP: specific specific GigE Vision CoaXPress from specific specific IMX492 Sony from specific specific TDI CCD: specific specific time-delay integration from specific specific 2K 4K 8K line scan from specific specific 400 kHz line rate from specific specific 500 MHz pixel clock from specific specific solar cell PCB wafer from specific specific 100 kV X-ray flat-panel from specific specific a-Si indirect CsI from specific specific 30×30 cm 2540 DPI); BASLER: 300+; OMRON: 500+.

Deep learning defect detection few-shot anomaly detection and transfer learning innovations; 3D structured light phase-shift fringe projection and photometric stereo innovations; semiconductor wafer inspection dark-field bright-field and electron-beam eBeam innovations; and embedded FPGA GPU vision processor pipeline innovations represent core machine vision patent domains: DEEP LEARNING INSPECTION PATENTS: COGNEX; KLA; APPLIED MATERIALS; OMRON: specific DL inspection innovations (specific specific anomaly detection: specific specific PatchCore memory bank from specific specific ResNet-50 backbone ImageNet from specific specific SPADE k-NN embedding from specific specific MVTec AD AUROC 0.98 from specific specific few-shot learning: specific specific Siamese network from specific specific 10-50 defect images from specific specific meta-learning MAML from specific specific defect segmentation: specific specific U-Net encoder skip from specific specific DeepLabV3+ atrous from specific specific PSPNet pyramid pooling from specific specific mIoU 92%+ from specific specific GAN synthetic defect: specific specific CycleGAN defect transfer from specific specific DCGAN texture from specific specific SSD YOLO: specific specific SSD MobileNetV2 from specific specific YOLOv8 300 fps GPU from specific specific tensorRT TensorFlow Lite from specific specific 6ms inference from specific specific Hailo-8 NPU 26 TOPS from specific specific 3D deep learning: specific specific PointNet PointNet++ from specific specific VoxelNet 3D box from specific specific BEV bird eye view KITTI from specific specific wafer inspection DL: specific specific ResNet-34 line scan TDI from specific specific pseudo-defect labeling from specific specific CMP ESD scratch particle 99.9%); STRUCTURED LIGHT 3D PATENTS: GOM; ZEISS; COGNEX 3D; KEYENCE; HEXAGON: specific 3D innovations (specific specific fringe projection: specific specific Gray code phase shift from specific specific 3-step 4-step sinusoidal from specific specific phase unwrap from specific specific 0.1 μm Z resolution from specific specific 10 μm absolute from specific specific 1M pts capture from specific specific photometric stereo: specific specific 4-direction LED from specific specific surface normal N xyz from specific specific Lambertian BRDF from specific specific roughness Ra 0.1 μm from specific specific structured light: specific specific DLP DMD 0.8 μm per fringe from specific specific 1024×768 LED projector from specific specific 30 fps 1M pts/s from specific specific stereovision: specific specific baseline 200 mm from specific specific stereo matching SAD NCC from specific specific disparity map 0.1 pixel sub-pixel from specific specific laser triangulation LJ: specific specific laser line 50 mW class 2M from specific specific receiver 5 MP CMOS from specific specific angle 20-30° from specific specific 0.5 μm Z 50 μm X from specific specific confocal chromatic: specific specific white light axial chromatism from specific specific 0.01 μm Z uncertainty from specific specific ISO 10360-8 traceable); SEMICONDUCTOR INSPECTION PATENTS: KLA; APPLIED MATERIALS; TOKYO ELECTRON; ASML: specific semicon inspection innovations (specific specific KLA 29xx wafer inspection: specific specific 193 nm DUV bright-field from specific specific dark-field oblique from specific specific 0.5μm sensitivity 65nm node from specific specific KLA eSL10 eBeam: specific specific 1kV-15kV electron from specific specific 3nm resolution from specific specific E-die to die from specific specific through-mask imaging from specific specific ASML HMI e-beam: specific specific multi-beam 1024 beams from specific specific 10nm resolution from specific specific 100 wafers/hr from specific specific OPC optical proximity: specific specific model-based SEM contour from specific specific 3nm MEEF from specific specific EUV inspection: specific specific 13.5 nm reflecting optics from specific specific attiZero actinic).

Coaxial darkfield ring dome illumination and multispectral LED wavelength selection innovations; camera calibration distortion correction photogrammetry and stereo calibration innovations; and vision-guided robot VGR eye-in-hand eye-to-hand calibration and 6-DOF bin-picking innovations represent additional machine vision patent domains: ILLUMINATION PATENTS: EFFILUX; CREE; COGNEX; OMRON: specific illumination innovations (specific specific coaxial back-illumination: specific specific beamsplitter 45° telecentered from specific specific 850 nm NIR LEDS from specific specific 625 nm red homogeneous from specific specific dome illumination: specific specific 150 mm dome Lambertian from specific specific aluminum oxide diffuser from specific specific ring light: specific specific 48 LED ring direct from specific specific 72° angle of incidence from specific specific low-angle grazing from specific specific darkfield: specific specific 0-5° oblique from specific specific surface defect highlight from specific specific side-light rail from specific specific multispectral: specific specific 365 nm UV fluorescing from specific specific 450 nm 530 nm 660 nm 850 nm from specific specific spectrograph hyperspectral pushbroom from specific specific 400-1000 nm 280 band from specific specific laser speckle illumination from specific specific structured light DLP from specific specific strobe sync 1 μs from specific specific LED driver constant current 0.1% ripple from specific specific telecentric lens: specific specific telecentric error < 0.01% from specific specific 0.1 × to 2 × magnification from specific specific distortion < 0.05% from specific specific bi-telecentric 0.01%); CALIBRATION PATENTS: OPENCV; COGNEX; PTB; NBS: specific calibration innovations (specific specific camera calibration: specific specific OpenCV Zhang 2000 from specific specific checkerboard 9×6 corners from specific specific Brown-Conrady distortion from specific specific k1 k2 k3 radial from specific specific p1 p2 tangential from specific specific focal length fx fy cx cy from specific specific 0.1 pixel RMS from specific specific stereo calibration: specific specific epipolar R T from specific specific RANSAC fundamental matrix from specific specific <0.5 pixel reprojection from specific specific photogrammetry: specific specific coded target retro from specific specific Aicon TRITOP from specific specific 0.5mm@3m Ax×10⁻⁵ m from specific specific hand-eye calibration: specific specific Tsai Lenz Shah method from specific specific AX=XB from specific specific 0.1 mm 0.1° accuracy); VGR PATENTS: ABB; FANUC; KUKA; UNIVERSAL ROBOTS: specific robot vision innovations (specific specific bin-picking: specific specific 3D vision 1M point cloud from specific specific OPC Open Protocol from specific specific GripML grasp pose from specific specific 90% pick rate from specific specific eye-to-hand: specific specific robot cell fixed camera from specific specific TCP offset transform from specific specific wrist camera eye-in-hand from specific specific DH Denavit-Hartenberg T from specific specific 3D-AOI solder: specific specific 5μm coplanarity from specific specific SPI solder paste inspection from specific specific solder bridge void lift from specific specific 3D AOI BTB SMT from specific specific 99.9% coverage from specific specific IPC-A-610 class 3 from specific specific vision with force: specific specific F/T sensor 80N 4Nm from specific specific compliance assembly from specific specific peg-in-hole 100 μm from specific specific guided by vision).

Machine vision inspection startup IP strategy must navigate Cognex PatMax geometric pattern matching and ViDi deep learning inspection patents (2,000+), Keyence 3D laser profile and chromatic confocal sensor patents (1,000+), Teledyne DALSA TDI CCD line scan and flat-panel X-ray camera patents (500+), KLA semiconductor wafer bright-field dark-field and eBeam inspection patents (2,000+), GOM Zeiss 3D fringe projection and photogrammetry patents (500+), and ABB FANUC vision-guided robot bin-picking and eye-in-hand calibration patents (500+); understand that Cognex holds the most significant 2D pattern matching and deep learning defect inspection IP, KLA holds semiconductor wafer inspection IP, and Keyence holds 3D profile sensor patent density in industrial; identify whitespace in novel anomaly detection with PatchCore or CFA on embedded GPU NPU (<10ms), novel multi-modal 2D+3D+hyperspectral fusion for food pharma textile inspection, novel EUV actinic reticle inspection beyond KLA (photomask defect <1 nm EUV 13.5 nm), and novel tactile+vision sensor fusion for robot assembly (1 mm force/torque 3D fusion): MACHINE VISION STARTUP IP STRATEGY: UNDERSTAND THE MACHINE VISION PATENT LANDSCAPE — COGNEX PATMAX AND KLA DUV INSPECTION HOLD BROAD FOUNDATIONAL IP: Cognex PatMax pattern matching and ViDi deep learning inspection patents and KLA 29xx/eSL10 bright-field eBeam semiconductor patents cover the two dominant vision inspection market segments — new entrants need novel deep learning architecture (few-shot anomaly detection <50 images, PatchCore embedded GPU), novel sensor modality (hyperspectral pushbroom 400-1000nm, LiDAR+camera fusion), or novel application domain (EV battery cell inspection, OLED panel display micro-LED defect, pharmaceutical blister pack); NOVEL ANOMALY DETECTION FEW-SHOT DEEP LEARNING AND MULTI-MODAL SENSOR FUSION ARE HIGHEST-VALUE LEAST-CONSOLIDATED IP: After Cognex ViDi ResNet YOLO 100-500 images training and KLA DUV 0.5 μm bright-field sensitivity, novel industrial anomaly detection with PatchCore memory bank and only 10 normal images training (no defect labels), and novel 2D+3D+hyperspectral fusion deep learning single-pass inspection represent less consolidated patent territory; SEMICONDUCTOR ADVANCED PACKAGING AND EV BATTERY INSPECTION CREATE NEW COMMERCIAL MACHINE VISION PATENT SPACE: HBM chip-on-chip 3D IC 3.5D/2.5D advanced packaging inspection (5μm interconnect bump void), micro-LED 40 μm pixel array transfer defect, and prismatic cylindrical EV cell separator electrode inspection represent high-value emerging machine vision IP — novel e-beam multi-beam inspection, novel phase contrast CT, and novel AI-native inspection recipe self-programming are whitespace; WHEN TO PATENT IN MACHINE VISION: NOVEL ALGORITHM WITH MEASURED DEFECT DETECTION RATE, FALSE POSITIVE RATE, AND THROUGHPUT: specific novel inspection system (specific specific sensor type + specific specific algorithm + specific specific defect size μm + specific specific detection rate % + specific specific false positive % + specific specific throughput ppm/hr) vs. specific Cognex ViDi 99.99% uptime 30ms cycle PatMax 1μm 0.01° or specific KLA 29xx 0.5μm sensitivity DUV 65nm node or specific LJ-X8900 0.1μm Z 3.2Mpts/s baseline — measured defect sensitivity μm, AUROC, mIoU, throughput ppm/hr, and false positive rate vs. Cognex/KLA/Keyence baseline is the critical machine vision IP metric; KEY FTO CHECKLIST: Cognex PatMax 1μm 0.01° occlusion; ViDi ResNet U-Net YOLO 100-500 images 99.99%; Keyence LJ-X8900 408nm 3.2Mpts 0.1μm Z; 3D-A7000 confocal 0.02μm Z; KLA 29xx DUV 193nm 0.5μm eBeam eSL10 3nm multi-beam; Teledyne TDI 400kHz linescan GigE; GOM Zeiss fringe phase-shift Gray-code 0.1μm Z; ABB FANUC bin-picking 90% 3D OPC T TCP; OpenCV Zhang distortion calibration stereo; MVTec AD PatchCore AUROC 0.98; U-Net DeepLabV3+ mIoU 92%; ASML HMI actinic EUV 13.5nm.