Combining where and what in change detection for unsupervised foreground learning in surveillance
作者：
Highlights：
• Build multi-appearance detectors for unknown and uncontrolled sequences in unsupervised manner.
• Global discriminative optimization based on latent SVM is able to build accurate multi-class detectors without pretrained detectors.
• From a noisy initialization (motion cues) learn position, scale and appearance of multiple foreground objects.
• Combine in an unsupervised way where (motion segmentation) and what (learning procedure) in change detection.
• Handle an unknown number of objects in an unconstrained scenario.
摘要
Highlights•Build multi-appearance detectors for unknown and uncontrolled sequences in unsupervised manner.•Global discriminative optimization based on latent SVM is able to build accurate multi-class detectors without pretrained detectors.•From a noisy initialization (motion cues) learn position, scale and appearance of multiple foreground objects.•Combine in an unsupervised way where (motion segmentation) and what (learning procedure) in change detection.•Handle an unknown number of objects in an unconstrained scenario.
论文关键词：Object detection,Unsupervised learning,Motion segmentation,Latent variables,Support vector machine,Multiple appearance models,Video surveillance
论文评审过程：Received 15 December 2013, Revised 19 September 2014, Accepted 29 September 2014, Available online 19 October 2014.
论文官网地址：https://doi.org/10.1016/j.patcog.2014.09.023
