The eyes of industry 4.0 - Vision intelligence and machine learning
summary
Distinguishes when deep learning helps versus when geometry wins: image classification, object detection and segmentation work well but require labeled data and roughly 20–700 images for training; position and angle measurements need sub-degree accuracy so classical 2D/3D pattern matching gives higher positional accuracy and handles 6‑DOF robot picks in the Lexington robot cell. Deep nets excel with overlapping or many classes and uncontrolled lighting — a warehouse shelf empty/occupied case replaces a costly 3D sensor with a 2D camera, saving money and improving reliability.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:50Typical deep learning applications
- 6:39Template matching workflow
- 7:39Object detection with deep learning
- 11:01Code reading - where deep learning meets traditional methods
- 13:52Vision Guided Assembly
- 15:05Defect classification - a typical deep learning application
- 16:09Identifying empty shelf slots
- 18:02Is Deep Leaming the holy grail of machine vision?
- 20:03Deep Learning complements the machine vision toolbox
related talks
14mComputer vision on edge devices
41mPractical Aspects of Applying Modern Computer Vision Technologies
27mComputer vision for automated quality inspection
13mEnhancing Object Detection & Avoidance Systems for Robots in Challenging Environments Using AI
41mDeep Learning to Solve Challenging Problems
46mImage Recognition with Deep Learning From prototype to production
40mMachines are Learning
28mChallenges & future trends of Deep Learning