Chasing birds with code
summary
A working system captures an RTSP camera feed on a Raspberry Pi + camera, pushes frames into Redis Streams + Gears + RedisAI, runs a YOLO model inside Redis to detect birds, and fires an HTTP callback to the Pi to play deterrent sounds. Tiny YOLO reaches about 40 fps on small hardware, switching to a fuller YOLO drops throughput to ~1 fps. Prototype runs reliably but has detection gaps, hardware limits (Pi 2 ~15 fps capture) and plans for GPU or parallel processing to improve accuracy and speed.
AI-written summary · about
chapters (19)
- 0:00Introduction to project
- 1:30Bird problem in Israel
- 3:16System architecture overview
- 5:00Why use Redis
- 6:05Redis Streams explained
- 7:33RedisGears and RedisAI
- 9:22HTTP server for Raspberry Pi
- 10:46YOLO model and image processing
- 11:55Capturing images from IP camera
- 13:20Bottle framework for HTTP server
- 14:23Hardware challenges with Raspberry Pi
- 15:47Model performance issues
- 17:32Real-time demo setup
- 20:01Live demo of human detection
- 21:50Performance and memory usage
- 22:33Q&A session part 1
- 24:02Q&A session part 2
- 25:25TensorRT and AI backends
- 26:21Data retention in Redis Streams
related talks
23mBirding with Python and Machine Learning
21mAI, Embedded System and Robotics
10mDeep Learning in the Wild: Embedded PyTorch for...
36m"Lights, camera, AI!" - Automated sports videography
30mWhen performance matters: running a real-time computer vision pipeline
26mFrom Software to Hardware: How Do I Track My Cat with JavaScript
37mUnstructured Data Processing with a Raspberry Pi AI Kit and Python
01h 00mRun your own IoT device in a snap