|
Post by Devonator on Dec 5, 2012 4:19:57 GMT -5
Person Recognition
Task
Detect that a person is standing in front of a camera Identify and label that person Remember that individual upon subsequent visits or distractions Identify that individual correctly with multiple people in view
Why
Get familiarized with OpenCV or OpenNI Important component to competition Evaluate limitations with Web-cam or Kinect
Task Limitations
Memorize only one person at a time.
Evaluation
Test Cases Re-identify person after occlusion Remember person out of frame Rotation and scale invariant Ignore other people in frame 90% success rate
|
|
|
Post by galotalp on Jan 5, 2013 17:51:47 GMT -5
How can our agent recognize people using the kinect? There are a few things that make people unique in this context, their height, their proportions, their color, and their faces. From an engineering perspective, we should conduct some analysis to see which of these factors is most useful to our agent in person recognition. However I'm sure we all know that the answer to that question is the 'face' factor. Unfortunately facial recognition is a harder to implement than 'limb-proportion recognition'. Fortunately, we're not going to have to implement it ourselves due to the wonders of stack overflow: stackoverflow.com/questions/953714/face-recognition-library
|
|