Accurate measurement and prediction of pellet size distributions are critically important for material processing because they are essential for model predictive control, real-time optimization, planning, scheduling and decision support of material production. Mechanical sieving is one of the traditional methods for pellet size measurement in industrial practice but cannot be applied in real-time fashion. Alternately, multivariate image analysis based pellet sizing methods may acquire the size information non-intrusively and thus can be implemented for on-line measurement in industrial applications. Nevertheless, the conventional multivariate image analysis based pellet sizing methods cannot effectively deal with the pellet overlapping effects in the still images, which may lead to inaccurate and unreliable measurements of size distributions. In our study, two novel video analysis based pellet sizing methods are proposed for measuring the pellet size distributions without any off-line and intrusive tests. The videos of free-falling pellets are first taken and then the free-falling tracks of pellets in video frames are analyzed through the two video analysis based pellet sizing approaches. In the first video analysis method, the Sobel edge detection strategy is adopted to identify and isolate the free-falling tracks in order to estimate the diameters of the corresponding pellets. For the second video analysis approach, the filtered gray-scale video frames are scanned row by row and then the particle diameters are estimated and predicted through the built Gaussian process regression (GPR) models and a fine designed counting rule so as to eliminate the overlapping effects of nickel pellets along the horizontal and vertical directions. The utility of these two video analysis based pellet sizing methods is demonstrated through the measurement and estimation of free-falling nickel pellets in two test videos. (C) 2014 Elsevier Ltd. All rights reserved.