0

I am writing opencv code to detect vehicles from a aerial platform. How many pixels do i need on the target to a) detect b) track and c) train my image analysis code? A technical source with how false positives/negatives vary with pixel resolution is preferred, but personal experience is fine. The pixel resolution required will depend on the image analysis technique I use. I havent decided on a method. It will probably be Haar, or HOG or a readlily available object detection algorithm.

user3290530
  • 1
  • 1
  • Seems to be a pretty broad request. Perhaps doing a bit of research to start and then come back when you have something more specific? – sous2817 Feb 09 '14 at 20:44

1 Answers1

0

I am using 90*90 resolution for training and i am getting good results. Actually my object contains considerable amount of area in this 90*90 image. So, it depends whether your feature detector is able or not to extract the important features from the image with the given resolution. I would suggest you to check first with few example with different size and go for higher resolution if you think that the required features cannot be extracted at the given resolution.

skm
  • 5,015
  • 8
  • 43
  • 104