Interactive Multiple Object Learning with Scanty Human Supervision [BoBot] [OpenCV Code]

preview_player
Показать описание
Interactive Multiple Object Learning with Scanty Human Supervision
M. Villamizar, A. Garrell, A. Sanfeliu and F. Moreno-Noguer
Computer Vision and Image Understanding (CVIU), 2016

Description:
This video shows online learning and detection results of multiple objects for robotics applications. The proposed method learns multiple objects on the fly using minimal human assistance. Specifically, the method integrates an online random ferns classifier that is progressively computed via human annotations with an uncertainty threshold that reduces the degree of human interventions.

Green rectangles indicate the output of proposed method, whereas magenta boxes are uncertain detection hypotheses that are labeled by the human as background samples. Blue boxes correspond to the ground truth.

Links:

Contact:
Michael Villamizar
Institut de Robòtica i Informática Industrial CSIC-UPC
Barcelona - Spain
Рекомендации по теме