Geometric Bounding Box Interpolation: An Alternative for Efficient Video Annotation

Link to published paper

Abstract

In order to reduce the cost of video labeling, instead on annotate every frame of a trajectory, usually only a sparse set of annotations are provided by the user, typically its end points, plus some intermediate key frames, interpolating the remaining annotations between key frames. While a number of video annotation tools have been proposed, some of then also freely available, bounding box interpolation are mainly based on image processing techniques, whose performance is highly dependent on image quality, occlusions, etc. In this work, we propose an alternative method to interpolate bounding box annotations, based on Cubic Splines and the geometric properties of the elements involved, rather than image processing techniques.

The algorithm proposed has been compared with other bounding box interpolation methods described in the literature, using a set of selected videos modeling different types of object and camera motions. The experiments carried out show that the accuracy when generating the interpolated bounding boxes is higher than the rest of the evaluated methods, specially when considering rigid objects. The main goal of this paper is related with the bounding box propagation step, believing that our design can be integrated seamlessly with any annotation tool already developed.



Annotation example
This trajectory has been labelled by providing only three annotations (white bounding boxes). Yellow bounding boxes are computed using the interpolation method implemented in this project.
(Flash Player needed for visualization)

NEW!!!!

Python
Download TrATVid

User Guide (PDF)

Any question, suggestion or technical request are welcome


Downloads
Source Code (and test) of the Evaluation Tool
All videos
Gallery
Videos
2 folders

Pedro Gil Jiménez. mail: pedro.gil at uah.es. University of Alcalá. Alcalá de Henares, Madrid, Spain.