- 學位論文
適用自定義資料集在教學環境下學生行為之及時偵測
Real-time Detector for Student Behavior in the academic environment
摘要
此篇論文目的是在教學環境中，使用影像辨識技術，辨別學生在上課中的行為，並根據精準教育中學習行為與學習環境，透過YOLOv4(You Only Look Once)這個網路架構，搭配Stanford 40 Actions行為資料集，時做出及時目標監控為導向的行為偵測器，並應用在大學課堂中。找出那些對於那些學習無益的行為，此系統會偵測學生的行為並呈現在螢幕上，以此將學生的上課表現呈現給教育者。 此外，YOLOv4是由以下四個部分組成:Input,Backbone:CSPDarknet53,Neck: SPP,Heads: Dense Prediction, Sparse Prediction，因此與前面幾個版本比起來有更高的準確率，以及更好的骨幹網路進行訓練集的特徵提取，因此在學習行為中，有些對學習無益的行為可將其標註並放進訓練集，並使偵測器可以呈現出學生的上課狀況給教學者看。 YOLOv4能夠應用在各種環境，例如：教學環境上，可以根據學生在教室中的行為(玩手機、喝水、寫筆記…)，訓練出一個適用於學習表現評估的偵測器或是在商場裡根據顧客行為，分析顧客對商品的購買意願。而本研究將此模型套用到教學環境中，讓教學者在螢幕上，顯示出現場學生的行為，並且能有效讓教學者了解到現場學生的專心程度。
並列摘要
The purpose of this paper is using the skill of image detection to detect the behaviors of students in the class,and base on the concept of Precision Education which needs the student’s behavior in the academic environment,Throught the network architecture of YOLOv4(You Only Look Once),training with the Stanford 40 behavior dataset, we built a behavior detector,and applies in the college class.To find those behaviors which are disbenefit to learning,this system will detect those behaviors of students and show them on the monitor,therefore the in-class performance can be shown to the educator. In addition,YOLOv4 is composed of following four parts:Input,Backbone:CSPDarknet53,Neck:SPP,Heads: Dense Prediction,SparsePrediction,therefore compared with the former versions YOLOv4 got higher accuracy and better backbone to do the features extraction fro, the datasets,therefore in learning behaviors,we can labeling those behaviors which are disbenefit to learinig,and built a new dataset,then the detector can showing the in class performance to the educator. YOLOv4 can be applies in different environments,forexample:in the academic environment,according to in class behavior(using phone,drinkingwater,take notes),we can train a detector of in-class performance or we could applie it in the business use,to analyzed the behaviors of guests,to tell the purchase intention.In this research we applies this architecture in academic environment,showing the behaviors of students on the monitor,letedcucators knew the concentration of students in real-time.