| 基于 YOLOv3 算法的教室学生检测与人数统计方法 |
| |
| 引用本文： | 沈守娟,郑广浩,彭译萱,王展青.基于 YOLOv3 算法的教室学生检测与人数统计方法[J].教育技术导刊,2009,19(9):78-83. |
| |
| 作者姓名： | 沈守娟 郑广浩 彭译萱 王展青 |
| |
| 作者单位： | 武汉理工大学 理学院，湖北 武汉 430070 |
| |
| 基金项目： | 国家级大学生创新创业训练计划项目（201910493067） |
| |
| 摘 要： | 教室环境中人群密度高、重叠部分较大的特点会给人数统计工作带来困难。基于深度学习 YOLOv3 目标检测算法对学生目标进行检测，并加入 Deep-Sort 算法为核心实时追踪方法，对 YOLOv3 算法检测到的学生目标进行一段时间的跟踪，从而克服传统视频人数统计方法中忽略视频上下帧关联信息的缺点，并且能更好地解决视频遮挡问题。目标检测方法中的损失函数用 tan 方损失函数代替原有交叉熵损失函数，跟踪算法中的卡尔曼滤波算法采用 Levenberg-Marquardt 对修正后的轨迹预测予以优化。最后，对该方法进行性能评价与对比实验，包括是否加入上下文信息对比以及与 SVM 目标检测算法作对比。实验结果表明，此方法在测试集中，加入上下文信息后准确率达 93.4%，召回率达 81.4%。对比 SVM，该方法在教室视频人数统计中准确率提升2.1%，召回率提升 8.9%。
|
| 关 键 词： | YOLOv3 算法 Deep-Sort 跟踪 人数统计 目标检测 损失函数 |
| 收稿时间： | 2020-02-01 |
Crowd Detection and Statistical Methods Based on YOLOv3 Algorithm in Classroom Scenes |
| |
| SHEN Shou-juan,PENG Yi-xuan,ZHENG Guang-hao,WANG Zhan-qing.Crowd Detection and Statistical Methods Based on YOLOv3 Algorithm in Classroom Scenes[J].Introduction of Educational Technology,2009,19(9):78-83. |
| |
| Authors: | SHEN Shou-juan PENG Yi-xuan ZHENG Guang-hao WANG Zhan-qing |
| |
| Institution: | Faculty of science，Wuhan University of Technology，Wuhan 430070，China |
| |
| Abstract: | The characteristics of high crowd density and large overlapping parts in classroom environment will bring difficulties to the work of population statistics. Based on the deep learning YOLOv3 target detection algorithm to test the students’goal，and join the Deep-Sort algorithm as the core of real-time tracking method，to track students target detected by YOLOv3 algorithm for a period of time，which can overcome the weakness that traditional video statistical methods ignore the video frame up and down in the correlation information，and can better solve the problem of video occlu-sions. The loss function in the target detection method replaces the original cross entropy loss function with the tangent square loss function，and the kalman filter algorithm in the tracking algorithm also adopts the trajectory prediction after levenberg-marquardt optimization correction. Finally，performance evaluation and comparison experi? ments are carried out for this method， including whether to add context information comparison and comparison with SVM target detec? tion algorithm. The experimental results show that the accuracy of this method is 93.4% and the recall rate is 81.4% after adding context information in the test set. Compared with SVM，the accuracy of this method in classroom video attendance statistics increased by 2.1%，and the recall rate increased by 8.9%. |
| |
| Keywords: | YOLOv3 algorithm Deep-Sort tracking crowd statistics target detection loss function |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
