A method of student classroom behavior recognition based on gesture information extraction

Abstract

The invention discloses a method for recognizing students' classroom behavior based on gesture information extraction, comprising the following steps: S1, model training, using known student classroom behavior images and class labels corresponding to the classroom behavior images to establish local feature recognition convolutional neural networks network; S2, using the local feature recognition convolutional neural network established in step S1 to recognize unknown students' classroom behavior images. The present invention uses posture information and local image information to identify students' classroom behavior images, effectively reducing the interference of information unrelated to behavior in the image, such as clothing color, body size, etc. Listening to lectures, looking around, sleeping, playing with mobile phones, taking notes, and reading books. Compared with the traditional image recognition method based on convolutional neural network, it can effectively improve the generalization accuracy of students' classroom behavior recognition.

Description

technical field [0001] The invention belongs to the technical field of computer vision and behavior recognition, in particular to a method for recognizing students' classroom behavior based on gesture information extraction. Background technique [0002] In traditional teaching classrooms, in addition to teaching knowledge, teachers are often responsible for maintaining classroom order. If the classroom order is chaotic, the level of students receiving knowledge is often unsatisfactory. With the continuous deepening of teaching informatization and intellectualization, in order to allow teachers to focus more on imparting professional knowledge to students, people will place their hopes on building an automated classroom teaching management system, how to accurately measure students' behavior in the classroom. identification becomes a challenging task. [0003] Common methods of student classroom behavior recognition include methods based on traditional machine learning and ...

AI Technical Summary

Problems solved by technology

Methods based on traditional machine learning need to manually extract appropriate features and use them to train classifiers for recognition. Such methods are difficult to design features and have low recognition accuracy

The deep learning methods emerging this year can carry out end-to-end training in a large amount of data environment, without the need for artificial design features, and the training process is more convenient, such as the method of student classroom behavior recognition based on deep learning proposed by Wei Yantao et al. (Reference: Yantao Wei, Daoying Qin, Jiamin Hu, Huang Yao, Yafei Shi. Classroom Behavior Recognition Based on Deep Learning. Modern Educational Technology, 2019,29(07):87-91.) Compared with traditional machine learning methods, such methods It reduces the difficulty of training and improves the recognition accuracy, but when the trained network is used to identify people who do not appear in the data set, the recognition accuracy will drop significantly

Images