Open AccessA practical algorithm for learning scene information from monocular video
Author(s) -
Lin Zhu,
Jie Zhou,
Jingyan Song,
Zhenlei Yan,
Quanquan Gu
Publication year - 2008
Publication title -
optics express
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.394
H-Index - 271
ISSN - 1094-4087
DOI - 10.1364/oe.16.001448
Subject(s) - computer science , pedestrian , computer vision , artificial intelligence , position (finance) , monocular , ground truth , pedestrian detection , field of view , field (mathematics) , remote sensing , geography , mathematics , archaeology , finance , economics , pure mathematics
The estimate of the scene information, such as the region of ground/non-ground, the relative depth of the ground and the unevenness of ground, is important for applications such as video surveillance, mapbuilding and etc. Previous research in this field is based on specific assumptions which are difficult to satisfy in practical situations. In this paper a practical algorithm is proposed to estimate the scene information in monocular video. With the pedestrian detection results for a period of time, the Pedestrian-Scene Map (PS Map), consisting of the average width of a pedestrian and occurrence probability of a pedestrian at each position of the scene, is learned by integrating the pedestrian samples with different sizes at different positions of the scene. Furthermore, the relative depth of ground region, the ground/non-ground region and the unevenness of ground can be measured with PS Map. Experimental results illustrated the proposed method's effectiveness with stationary uncalibrated camera for unconstrained environment.