stanCode_AI

Pedestrian crossing aid by machine learning

Introduction

https://github.com/lily71903/stanCode_AI_Project/issues

When crossing intersection...

• It is hard for them to perform safety evaluation
• In Taipei city, only 6.9% (185/2679) of intersections have audio-assisted pedestrian signals

Current Researches

Convolution neural network (CNN) based traffic light detection and recognition Current application:

• Self-driving vehicles
• Driving assistance

Our Proposal

A vision-based model for detecting the real-time intersection condition

1. Traffic light recognition
2. Crosswalk clearance
3. Estimated time of crossing

Methodology

Dataset & Data-preprocessing

• Self-constructed, 1148 in total
• Images of intersections in both Taipei city and Taoyuan city
• Raw data to trainable data: label → convert file → resize(1024*1024), crop, rotate → rename → create subsets
• Allocate our data into training, validation, and testing set by stratification in (90:5:5 ratio)

Results