Abstract
Statistical analysis is widely used for moving object detection from video sequences, and generally it assumes that the recent segmentations are the desirable samples. However, we think the recent segmentations are not desirable for the detection of deformable objects, such as a pedestrian. We present an object detection framework that is designed to select the most desirable samples from all historical segmentations for statistical analysis. Central to this algorithm is that the shape evolution of the deformable object is learned from historical segmentations by an autoregressive model. Based on the learned model, the shape of the moving object in a current frame is predicted. Those historical segmentations that show similar shape to the predicted shape are selected for statistical analysis instead of the recent segmentations. Definitive experiments demonstrate the performance of the proposed method.
© 2009 Optical Society of America
Full Article | PDF ArticleMore Like This
Sooyeong Kwak, Guntae Bae, and Hyeran Byun
J. Opt. Soc. Am. A 27(2) 180-187 (2010)
Sara Memar, Riadh Ksantini, and Boubakeur Boufama
J. Opt. Soc. Am. A 33(4) 648-662 (2016)
Xuan Zhu, Chao Zhang, JiaPing Xue, ZhenPeng Guo, RongZhi Wang, and YuYing Jin
J. Opt. Soc. Am. A 36(9) 1495-1504 (2019)