Blurred Image Detection from a Humanoid Generated Video Sequence

Abstract

This thesis presents a method for classifying the images acquired by a humanoid robot. The blurred images are detected and eliminated. The clear images are used to create a 3D model of the object shown in the images.

The 3D shape reconstruction from a humanoid generated video sequence project deal with the development of a strategy to estimate the geometry of an interesting object from a monocular video sequence acquired by a walking humanoid robot. In order to generate the 3D model of the object, firstly, blurred image must be eliminated. I have collaborated in this preprocessing step in which the final result is a set of images that contains the object without blur. First, the presence of blur is detected by the calculus of gradient magnitude. Second, a gradient histogram is displayed with twenty bins and the latest ten bins are added to classify the images. Finally, an image is considered as clear when the addition of the bins is bigger than 0.5% of the contour pixels.

El proyecto "3D Shape Reconstruction from a Humanoid Generated Video Sequence" trata de desarrollar una estrategia para estimar la geometría de un objeto a partir de una secuencia de video grabada por un robot humanoide. Para generar este model 3D del objeto, primero hay que eliminar las imágenes borrosas. Yo he colaborado en esta fase de pre-procesado en el que se seleccionan las imágenes que contienen al objeto nítido. Para ello, se detecta la presencia de borrosidad a través del cálculo de la magnitud del gradiente de la imagen. En segundo lugar, se construye el histograma usando veinte barras y se usa la suma de las últimas diez barras para clasificar las imágenes. Finalmente, se considera que una imagen está nítida cuando esta suma es mayor que el 0.5% de los píxeles de contorno.