Evaluation of motion estimation procedure for low-frame-rate aerial images

In an attempt to improve the quality of displacement estimations between frames in low-frame-rate aerial sequences, a displacement estimation algorithm related to frame-differencing techniques combined with an affine model of motion was tested. The algorithm follows the development of Lenz (1984) in which points on displaced gray-level edges are used to estimate the local apparent horizontal and vertical displacements with the subsequent global motion parameters for translation, rotation, and zoom calculated through linear regression. Performance measurements were made with real video sequences. Effects of scene content and flight parameters were observed along with effects of frame transmission rate, image coding procedures, and channel errors for simulated transmission. The algorithm worked well, producing correct matches about 90% of the time, and was largely unaffected by image coding procedures or channel errors