A comparison of the performance of image motion analysis algorithms operating on low signal to noise ratio images

Image motion analysis algorithms can be employed to generate the 2-D velocity, the direction of the 3-D velocity, the rotation, and/or the relative depth of objects in a sequence of images. Most of these algorithms are designed to operate on high-signal-to-noise-ratio (SNR) images (typical of optical and infrared images). The applicability of these algorithms to estimation for low-SNR images (typical of radar and sonar images) is addressed. Motion analysis algorithms can be segregated into two major categories: optical flow and feature-based algorithms. The effects of noise on the performance of a number of optical-flow-based algorithms are evaluated using Monte Carlo computer simulation on a set of standardized image sequences. The effects of feature coordinate perturbations on the performance of a feature-based algorithm are evaluated using Monte Carlo computer simulation. The 1-D FFT (fast Fourier transform) algorithm was found to yield the best results in the presence of noise. This algorithm is simple but has difficulty when multiple moving objects are present in the scene