Motion Direction Detection from Segmentation by LIAC, and Tracking by Centroid Trajectory Calculation

Motion information can form the basis of predictions about time-toimpact and the trajectories of objects moving through a scene. Firstly, a model that incorporates accumulative computation and lateral interaction is presented. By means of the lateral interaction in accumulative computation (LIAC) of each element with its neighbours, the model is able to segment moving objects present in an indefinite sequence of images. In a further step, moving objects are tracked using a centroid-based trajectory calculation. 1 Motion Direction Detection Motion plays an important role in our visual understanding of the surrounding environment [1]. Visual motion can aid in the detection of shape [2], provide information as to the relative depth of moving objects [3], and give clues about the material properties of moving objects, such as the rigidity and transparency [4]. Motion information can also form the basis of predictions about time-to-impact and the trajectories of objects moving through a scene [5]. This paper introduces a novel method for motion direction detection based on segmentation by lateral interaction in accumulative computation (LIAC) and tracking by centroid trajectory calculation. 1.1 Segmentation from LIAC The aim of segmentation step is firstly to determine in what grey level stripe a given element (x,y) falls. Let GLS (x,y,t) be the grey level stripe of image pixel (x,y) at time t and n the total number of grey level stripes. [ ] 256 , 1 , 256 ) , , ( ∈ = n n t y x GLS (1) Lateral interaction in accumulative computation is capable of modelling the motion on the image, starting from the pixel grey level stripe and the element state or permanence value. There are as many permanence values for a given element as grey level stripes. At each time instant t, the permanence value is obtained in two steps. (1) A charge or discharge due to the motion detection, that's to say, due to a change in the grey level stripe, and, (2) a re-charge due to the lateral interaction on the partially Fernández-Caballero A. (2005). Motion Direction Detection from Segmentation by LIAC, and Tracking by Centroid Trajectory Calculation. In Proceedings of the 5th International Workshop on Pattern Recognition in Information Systems, pages 213-218 Copyright c © SciTePress charged elements that are directly or indirectly connected to maximally charged elements. The charge or discharge behaviour of the permanence memory is explained next. (a) All permanence values not associated to grey level stripe k are completely discharged down to value vdis. (b) If the pixel associated to the element is enclosed in grey level stripe k, we are in front of two different possibilities. (b.1) If the pixel was not enclosed in grey level stripe k in time t-1, permanence memory is completely charged up to the maximum value vsat, or, (b.2) if the pixel was previously enclosed in grey level stripe k in time t-1, permanence memory is applied a decrement of value vdm (discharge value due to motion detection), down to a minimum of vdis.