Friday, December 1, 2006

Welcome!

Hello, this blog will be created to track the progress of a research project for CSE 190-A: Projeccts in Vision and Learning. The course is taught by Serge Belongie at UCSD. My project is titled "Vision Based Traffic Light Triggering." As the title suggests, I want to investigate how accurate we can detect motorbikes using a camera and image processing algorithms. Click here for the proposal.

I thought of this idea because usually at night time when I was riding my motorbike and there's no cars. This means that the light didn't get triggered for me.

The problem lies with the traffic light sensor. Traffic lights work by an inductive loop to detect a large presence. Motorcycles tend not to set these off.

Solutions:
There are a couple workarounds but each with their own faults:
  • a large magnet(maybe a hard drive?) taped to the motorcycle
    Not tested, not everyone will buy a magnet to stick on a bike.


  • running the red light
    BAD! Can get you in trouble with the man or killed by an SUV.

  • get off the bike and run to the nearest crosswalk button.
    Only works on going straight.



  • Abstract
    Current traffic signal light triggers do not work well with motorbikes. I plan
    to develop an algorithm or improve upon existing algorithms to classify
    motorbike presence at a traffic light with the main goal of improving safety
    for motorcyclists.