Motion estimation of glass eels by differential methods

In computer vision and image processing, motion estimation is of increasing interest because of the large number of applications: object tracking (military, video-surveillance, robotics), complex behavioral analysis (modeling of human body motions, meteorology), medical analysis (cardiac contraction follow-up, infarction detection) [1][2]. In biology, tracking the motion of animals sometimes poses technical problems, related to the characteristics of species and stages of development. For example, the European glass eel(Anguilla anguilla) has a complex life cycle, with reproduction in the sea of Sargasso, a larval phase that crosses the Atlantic Ocean and a juvenile stage, the glass eel, which goes up the estuaries to grow in the river [3]. To study the estuarine migration of glass eel, it is possible to reproduce the tidal currents in the laboratory and observe the swimming behavior of individuals [4]. The major difficulties concern the animal itself, which is transparent, and moves mainly at night or at very low light intensity. To follow glass eels, each individual is tagged with VIE Tag (Visible Implant Elastomer) [4]. This marking consists in implanting under the skin a tip of colored elastomer, visible under UV. Tracking individuals is done on video recordings but it is a tedious job because currently not automated. The parameters that interest biologists are mainly the motion direction of glass eels (with or against the current) and their velocity. Any measure to assess energy expenditure is also sought, as glass eels do not eat during migration, and their energy status could play an important role in the migration potential. In this work, we have chosen differential methods for their many advantages. These methods are at first robust and precise, while being easy to implement. Because of its differential nature, the optical flow equation also allows a sub-pixellic estimation of the motion [5,6]. The advantages of these methods are: firstly, robustness and precision, the equation of the optical flow, because of its differential nature, allows a sub-pixellic estimation of the movement. The measurement of the motion requires only a local calculation of the spatio-temporal derivatives of the sequence as explained in a recent study [7]. The main disadvantage of differential methods is their foundation based on constant light intensity assumptions, and small displacements. During larger trips, however, it is possible to solve the problem by multi-resolution iterative approaches. In this case, we do not only process the image sequence at its acquisition resolution, but we build from each frame, a pyramid of images successively filtered and subsampled [8] [9]. The relevance of such an approach requires, however, that the images at the coarsest levels of the pyramid should not be too degraded by the lowering of the resolution. The purpose of this preliminary work is to use differential methods to estimate the motion of glass eel, determine the direction of their motion and assess their velocity. The article is structured as follows: section 2 presents the experimental context of the study. Then we recall the two motion estimation algorithms, namely Lucas and Kanade algorithm and briefly H&S algo. Section 4 describes the proposed processing approach. Then we present the results. Finally, we end with a conclusion and perspective.