Video demo of object re-detection
This video demo presents the results of the object re-detection algorithm on a video from the documentary scenario (Sound & Vision; Tussen Kunst & Kitsch, AVRO). Object re-detection aims at finding occurrences of specific objects in a single video or a collection of still images and videos. The algorithm takes as input a picture (query image) of a manually specified object of interest by the user, who marks this object on one frame of the video with a bounding box. Then, this picture is compared against consecutive or non-consecutive frames of the video and the instances of the depicted object are automatically detected and marked with a bounding box. In this video demo, the detected re-occurrences of the object of interest are indicated by a green rectangle around them. The object re-detection algorithm is robust against a range of scale and rotation operations and partial occlusion. However, in some cases, extremely different viewing conditions (due to major modifications in scale and/or rotation), under which the object's re-appearance takes place, lead to significant change of the visual information, and thus detection failure.