Providing Better Feedback in Real-time Object Detection Apps
summary
Real-time business-card scanning app Wantedly People detects multiple cards but initial overlays jump and disappear because detections arrive intermittently; adding stable tracking IDs lets the UI animate each card’s contour from its previous frame to the current one. OpenCV trackers cause performance issues on some iOS devices, so a lightweight custom tracker that combines three cheap heuristics runs fast and keeps labels consistent. The team adds bubbling animations to each layer, improving user understanding, and links a sample app using CIFaceFeature on GitHub.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:37Object Detection
- 0:54Object Detection Demo
- 1:06Outline
- 2:47Object Frequency
- 3:04Multiple Objects
- 3:28Tracking
- 3:43Tracking Logic
- 4:00Codfish
- 4:41Load Map
- 5:02CI Face
- 5:23Summary
related talks
46mRealtime Face Detection with Firebase ML Kit
21mLeveraging Python for Real-Time Image processing
31mBuilding Camera Features in a Snap
26mIs This Your Card? Computer Vision for Playing Card Recognition
33mSee More, Be More
37mSeeing is Believing: Mobile Vision API Deep Dive
30mMobile applications that can see
12mCascade models in computer vision to boost accuracy and performance