Detection: how many pixels are required?

Abstract

In this work, we studied how detection range of a small target changes as a function of the number of defined target pixels. The intuitively simplest method to look for a target in IR images is to look for the “hottest” (highest intensity) pixel. Image noise or sun glint may cause false detection. To build in some robustness against such detractions, one considers looking for the hottest group (blob) of contiguous pixels. A blob could be any shape; here we consider square blobs, 1x1, 2x2, 3x3, etc. pixels. One expects the average blob intensity to decrease with blob size. On the other hand, the noisiness of the background blobs also decreases with blob size. The net result is that initially, for small blob sizes, the detection range increases before falling again for larger blob sizes. We demonstrate this by analyzing IR recordings of a small vessel sailing outbound until it “disappears”. We develop a simple model that supports these observations. The model is based on a synthetically generated sequence of images of a receding target, uses a basic sensor characteristic, and Johnson’s detection criterion.