Classroom human body posture real-time detection method based on multi-scale features

Abstract

The invention relates to a classroom human body posture real-time detection method based on multi-scale features, and belongs to the technical field of image processing and computer vision. The methodcomprises the following steps: based on an SSD network of single-stage target detection, performing data enhancement on classroom images through partitioning processing, extracting basic features ofthe images through a VGG16 network, and extracting cascaded multi-scale feature maps through an RFB module; down-sampling the original image through a multi-scale feature enhancement branch, sending the down-sampled image to a multi-scale feature enhancement module to learn multi-scale information, and fusing the down-sampled image with the cascaded multi-scale feature image in a point multiplication mode; complementarily fusing the fine-grained features of the shallow layer and the semantic features of the high layer through a feature adaptive fusion module so as to improve the ability of thenetwork to identify similar posture categories, and finally classifying and regressing the adaptively fused feature map; on the premise of ensuring real-time performance, the problem of human body posture detection in a classroom scene with large human body density and serious shielding is solved.

Description

technical field [0001] The invention belongs to the technical field of image processing and computer vision, and relates to a real-time detection method of human body posture in classrooms based on multi-scale features. Background technique [0002] The attitude information of students is of great significance to teaching management and teaching evaluation. If a teacher's class is successful, almost all students in the class will sit up and listen carefully, or stand up to answer questions and interact with the teacher. If a class is not enough to arouse students' interest, usually students will play on their phones or sleep on their desks. In addition, in the examination room, we can judge whether the student is suspected of cheating by whether the student stands up or lies on the table. However, the management and evaluation of classrooms and examination rooms are usually completed by manual visits and random inspections by supervisors, which is very time-consuming and i...

AI Technical Summary

Problems solved by technology

However, for the target in the classroom scene, the crowd is very dense, and the occlusion of the front and rear rows and the lower body are generally more serious. Due to the difference in the imaging distance of the surveillance camera, the resolution of the target in the front and rear of the classroom is quite different.

In the classroom scene, the reliability of the existing key point-based pose estimation method is low, so the method of target detection is more suitable for solving the human body pose state analysis task in the classroom scene

Images