Detection of characteristic eye points under challenging light conditions is a non-trivial task. Difficulty of this task increase even more when low resolution images are used. This article introduces an adaptive solution for detection of characteristic eye points in mentioned conditions. First, light normalization algorithm is performed. Then, face and eye detection followed by determination of eye status (namely: open or closed) is done. Finally, an adaptive method for pupil and eye corners detection is developed by comparing some existing methods. Experimental results show the outperformance of the proposed method.
Novel motion trajectory representations are introduced for efficient storage, transmission and search. Bit size reducing is obtained by specialization of motion trajectory temporal models to each spatial coordinate contrary to MPEG-7 approach where temporal models are the same for all spatial dimensions. On trajectories of MPEG-7 experimentation model (XM) the average bit gain is about 17%. For motion trajectory matching in search applications, the proposed similarity measure is based on exact integral formula for piecewise polynomial representation. This approach is not only more accurate but on average at least one order of magnitude faster than vector distance for discrete dense interpolation of trajectories, which is proposed in MPEG-7 XM. For long or complex shape trajectories matching in large databases can be too slow. Therefore we propose two acceleration techniques reducing significantly the number of trajectory comparisons: distance of trajectory centroids and distance of trajectory dispersion vectors. Using both techniques allows to skip about 95% trajectory comparisons when they are stored in X-tree or M-tree.