基于YOLOv3的教室人数检测

为了提高教室人数检测精度识别出教室的正确状态,克服学生上课姿态,座位远近、面部遮挡、尺度较小等因素而造成传统人脸检测技术识别精度较低,无法满足实际需求的问题。实验采用了如今热门的实时目标检测算法——YOLOv3(YOU ONLY LOOK ONCE v3),并根据教室这一特定场景采集大量图像信息样本,对其进行标注制成训练集。通过多次训练并不断调整参数,得出最佳的教室人数检测模型.实验结果表明,在1000张测试集图片样本中能得到90%以上的准确率,可以识别出教室人数并准确判断出教室状态,验证了基于YOLOv3的教室人数检测方法的有效性。

In order to improve the detection accuracy of the number of students in the classroom and recognize the correct state of the classroom,overcome the problems of students'classroom postures,seat distances,face occlusions,small scales and other factors which easily result in the low-accuracy recognition of traditional face detection technology,thereby can not meet the actual needs.In the experiment,we used the popular real-time target detection algorithm-YOLOv3(YOU ONLY LOOK ONCE v3).A large number of image information samples are collected according to the specific scene of the classroom and labeled to make a training set.By training many times and adjusting the parameters continuously,we got the best number detection model of classroom.The experimental results show that the accuracy rate of more than 90%can be obtained in 1000 test set picture samples,and the number of students in the classroom can be identified and the state of classroom can be accurately determined,which verifies the effectiveness of the number detection method based on YOLOv3.