葛仁彦-机器视觉的自适应算法在工业自动化检测中的应用.pptVIP

Automatic SL Inspection EquipmentNanovision? Contents 目录1 Background 背景资料2 Adaptive Algorithm Introduction 自适应算法简介3Adaptive Filter 自适应滤波4 Adaptive Pattern Restoration 自适应模式恢复5 Adaptive Pattern Matching 自适应模式匹配6 Typical Application Case 典型应用案例7 Future Work 将来的工作Nanovision? 1. Background 背景资料Industrial vision systems require expertise in multiple fields as they are compositions of differenttechnologiesArchitectures? PC + CameraSensors? Embedded Vision? Smart Cameras? PC + Smart Camera? CMOS/CCD? OtherOptics? Lenses? Prisms Beam Splitters? Color FiltersData CaptureMachine VisionRobitic GuidanceFactory Automation? Analog? Digital? PolarizersSpectrum / IlluminationProcessors?FPGAInterfaces? Visible? Infrared (IR)? Ultraviolet (UV) ? Other? LED? Halogen? Gigabit Ethernet (GigE)? Firewire (1394b)? CameraLink? USB 2.0 - 3.0? Other?DSP(Digital Signal Processor)?RISC(Reconfigurable Instruction SetComputer)?Frame GrabbersSource：Frost Sullivan，2010Nanovision? 1. Background?背景资料Current Algorithms method for Industrial Parts/Surface Inspection目前在工业部件和表面检测中的常用算法CAD Based PatternMatching与CAD设计图案匹配Design Rule CheckGolden Template MatchingDRC金色模板匹配设计规则检查Artificial Neural NetworkFeature based Compare人工神经网路基于特征的比较Nanovision? 1. Background 背景资料? 目前算法存在的问题? CAD?Based?Pattern?Matching?（与CAD设计图案匹配）? CAD模板图像化过程比较繁琐? CAD模板与实际采集图像很难配准，影像采集根据应用不同有不同程度的变形? Golden?Template?Matching（金色模板匹配 ）? 需要很多良品制作金色模板，没有统一性? 受影像采集过程中运动等影响，模板匹配会有局部偏差? Design?Rule?Check?–DRC（设计规则检查）? DRC需要知道被检测物品的设计规则，需要经验值? DRC只能检测非常微小的局部缺陷，有局限性? Artificial?Neural?Network （人工神经网络）? 学习矩阵需要大量的样本? 计算量比较大，计算收敛慢，对在线实时检测需要FPGA或DSP的加速计算? Feature?based?Compare?（基于特征的比较）? 特征点的提取计算量比较大? 被检测物品的特征很难量化，有的物品特征不是很明显Nanovision? 2. Adaptive?Algorithm?Introduction?自适应算法简介Adaptive Background normalization自适应背景归一化Adaptive PatternRestoration自适应模式恢复Adaptive Filter自适应滤波Adaptive AlgorithmsMachine VisionAdaptive Pattern Matching自适应模式匹配Nanovision? 3. Adaptive Filter 自适应滤波?自适应滤波存在于信号处理、控制、图像处理等许多不同领域，它是一种智能更有针对性的滤波方法，通常用于去噪。?自适应滤波是近年以来发展起来的一种最佳滤波方法。它是