| 摘要 |
The invention relates to a method for bounding an object in a video sequence F x,y,t . The method comprises obtaining a subset of pixels located in the object to annotate, in each frame of the video sequence. Spatio-temporal slicing is performed on the video sequence F x,y,t , centered on the obtained subsets of pixels, resulting in a first image F y,t obtained by an horizontal concatenation of first slices, comprising the obtained subsets of pixels, and resulting in a second image F x,t obtained by a vertical concatenation of second slices. A trajectory of the obtained subsets of pixels is displayed on both the first F y,t and second F x,t image. By using contour detection methods, a first and a second boundary are obtained on both the first F y,t and second F x,t image , around the trajectory of the obtained subsets of pixels. A bounding form around the object to annotate is obtained out of four points in each frame of the video sequence, wherein the coordinates of the four points of a frame t are obtained from the coordinates of the points located in the first and second boundary of the first and second image for that frame t. Advantageously, the bounding form is a rectangle drawn out of the four points, or an ellipse inscribed in that rectangle, or an ellipse comprising the four points. |
