Recognition and Localization of Parking Signs using Deep Learning
summary
Uses Google Street View images, cleans and crowdsources labels via Figure Eight to produce 3,000 annotations, trains a sliding‑window SVM baseline, then fine‑tunes a YOLO model (pretrained on ImageNet) for detection — YOLO (160 epochs, 19 conv layers) shows high sensitivity — and experiments with SSD plus active learning 1% labeled where low‑confidence examples get sent to the crowd. Signs are localized by triangulating multiple viewpoints to map rules to curb segments; model works well in San Francisco and generalizes to NY/LA but needs city‑specific fine‑tuning and OCR for rule extraction.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:35Re-designing parking rules by Nikki Sylianteng
- 2:59Parking notifications
- 4:18Structuring the data
- 4:41Developing computer vision models
- 5:29Making sense of a messy world
- 5:52Data Collection and Annotation
- 7:31Sliding window detector
- 8:20Improving the results
- 8:58Parking sign localization
- 9:31Evaluations
- 10:25Deep learning object detector
- 12:32Predictions
- 12:49Challenging case
- 13:43Active learning
- 14:54SSD results
- 15:16How well does the model work on new data?
- 15:58Future work
- 16:54Final Remarks
- 18:12Solving parking problems
related talks
01h 14mFoundations of Object Detection from an Expert's Overview
46mML Use Cases at Think Big Analytics
45mWhat if AI were the solution to understanding sign language?
29mRecognizing Traffic Lights With Deep Learning
26mSelf Driving cars with Opencv
07mFinding parking spots with Custom Vision and IoT
04mYOLO in R
10mMapping sidewalks and street furniture with AI and computer vision