摘要
针对单一的特征提取算法在运动员检测中还存在漏检较多的问题,本文提出了一种中层特征块分类的运动视频运动员检测模型。首先采用中层特征块作为描述运动员的特征,然后采用SLIC算法进行超像素分割,利用像素的CIELAB颜色空间和XY空间坐标构建像素的5维特征,最后采用高斯分量的全协方差混合高斯模型建立前景背景像素描述模型,提高检测精度。仿真实验结果表明,本文提出的改进模型,相比较HOG算法和SVM算法,检测结果更准确的表示了运动员区域。
Aiming at the problem that the single feature extraction algorithm has more problems in the detection of the athletes, this paper proposes a sports video player detection model with middle feature block classification. Firstly, the mid-level feature block is used to describe the characteristics of the athlete. Then, the SLIC algorithm is used to divide the pixel, and the 5-D feature of the pixel is constructed by using the CIELAB color space and the XY spatial coordinates. Finally, the Gaussian component is used to establish the foreground Background pixel description model to improve the detection accuracy. The simulation results show that the improved model proposed in this paper compares the HOG algorithm and SVM algorithm, and the detection result shows the athlete area more accurately.
出处
《科技通报》
北大核心
2017年第12期137-140,共4页
Bulletin of Science and Technology
基金
江西省教改课题(课题编号:JXJG-14-49-8)
关键词
特征提取
运动视频
中层特征块分类
运动员检测
超像素分割
全协方差描述
feature extraction
motion video
middle feature block classification
athlete detection
super pixel segmentation
total covariance description
