摘要
研究一种基于OpenCV和Haar特征检测固定区域图像中人数的方法。通过选取大量含有人的Haar特征的样本图片,利用OpenCV训练出分类器,并通过实验深入分析选取的样本对分类器识别性能(即检测效果)的影响,最终选取人的头肩部上半身样本训练分类器,并在1 500张640pix×480pix待检图像(共计人数17 294人)的检测实验中达到93.9%的识别准确率,平均检测时间小于323ms。
A method of detecting people number of image in fixed place with OpenCV and Haarlike classifier is researched.By selecting a large number of sample images containing human Haar-like features,a classifier is trained by using OpenCV.Through the experiment,finally the upper-body samples are chosen to train the classifier,and it is used to recognize 1 500 images with 640 pix×480 pix and totally contains 17 294 people.The accuracy rate can reach 93.9 percent,and the average time is less than 323 ms.
出处
《辽宁科技大学学报》
CAS
2011年第4期384-388,共5页
Journal of University of Science and Technology Liaoning
