A foreground detection model for double-trapezium cylinder codebooks (DTCC) is proposed to solve the problem that there exists shadow and highlight in foreground detection in intelligent visual surveillance. First an erected trapezium cylinder structure for shadow detection is constructed in the lower part of the model based on the characteristics that the shadow is less bright than the background and the chromaticity of shadow is different from that of the background. Then an inverted trapezium cylinder structure for highlight detection is constructed in the upper part of the model based on the characteristics that the highlight is brighter than the background and the chromaticity of highlight is different from the background. The DTCC model is formulated by finally constructing a main background area in the middle part of the model using a cylinder structure. Experimental results show that the average frame rate of the DTCC model is about 15. 61 frames per second and the real-time performance is higher than that of iGMMXBM and HC3 algorithms with 76.05%, 24.60% and 41.09%, respectively. The average shadow suppression rate and average highlight suppression rate of the DTCC model are about 70. 61% and 83.41%, respectively, and both the performance are better than those of the compared algorithms.%为解决智能视觉监控中前景检测存在阴影和高亮噪声影响的问题,提出一种双梯柱体码本(DTCC)前景检测模型.首先,根据阴影与背景相比亮度较低且色度不同的特点,将模型下部构造成正立的梯柱体阴影检测区；随后,根据高亮噪声与背景相比亮度较高、色度不同的特点,将模型上部构造成倒立的梯柱体高亮噪声检测区域；最后,模型中部采用圆柱体构造成主体背景区域,最终形成DTCC模型.实验结果表明:DTCC模型具有更高的实时性,平均帧率约为15.61帧/s,比iGMM模型、CBM模型和HC3模型的帧率分别高出76.05％、24.60％和41.09％；具有更优的阴影和高亮噪声消除性能,平均阴影和高亮噪声消除率分别约为70.61％和83.41％,均高于对比模型.
展开▼
