Real-time Face Recognition With Microsoft Cognitive Services
summary
Real-time face recognition uses Azure Face API for identification but relies on local pre-checks to stay affordable: crop images to ~15KB, batch up to 64 faces per image and identify up to 10 per call, cache group members instead of fetching each person, and only send a request when the face count changes rather than every frame. That workflow drops costs from about $70/day to under $1/week. Add Kinect or Intel RealSense for infrared, depth and skeleton tracking and on‑prem containers when data cannot leave the network.
AI-written summary · about
chapters (19)
- 0:00Intro
- 2:43Why Cognitive Services?
- 5:37Challenges
- 7:01Face API tiers
- 10:21Web API calls for 3 people
- 11:34What about streaming?
- 11:46Naive implementation (cost)
- 12:46Simple cost reductions
- 14:16Offline face detection libraries
- 15:45Tracking faces
- 17:25Detect moving faces
- 18:31Tracking number of faces (code)
- 21:26Additional tips #5
- 24:33Additional tips #6
- 27:01Low light environment - web cam
- 27:19Low light environment - Kinect v2
- 31:44Infrared camera
- 35:17IR/depth cameras with Cognitive Services
- 38:46Recap
related talks
43mFace It: How Azure Face Service Can Help You Recognize and Identify Faces
01h 04mBuilding and Training your own Custom Image Recognition AI
01h 00mA lap around Microsoft Cognitive Services Vision API
55mEasily Add Artificial Intelligence To Your Apps Using Machine Learning
01h 03mEasier AI allows everyone to build smart apps
47mBeyond the Usual Suspects: Emerging Cognitive Services
01h 05mMicrosoft Cognitive Services: Making AI Easy
44mThe age of applied AI