基于深度学习的实时监控图像中考生异常行为自动识别算法

An automatic recognition algorithm for abnormal behavior of candidates in real-time monitoring images based on deep learning

基于图像处理技术,提出了一种自动、准确地识别出考试中异常行为的算法,旨在维护考试的公平性和提高监考的效率,并进一步规范考场纪律。首先,采用Mosaic技术增强视频帧图像,从而对含小目标的图像集进行扩展。接着,利用改进的YOLOv5s目标检测算法对考生考试行为状态进行人体检测框的定位,快速并准确地识别出考生的人体位置。然后,采用SimplePose对考生人体的关键点进行定位,进一步精确地描述人体的姿态并更准确地识别出异常行为;另外,SimplePose具有轻量级的特点,可以快速处理图像并提取关键点信息,确保实时性。最后,利用ConvNeXt图像分类器对人体关键点图像进行分类,增强了模型的鲁棒性和稳定性。实验结果表明,所提出的实时检测方法对考试环境中考生作弊行为识别具有快速的检测性能,并且识别准确率达到92.1%,提高了考场监考的效率并维护了考试公平。

With the construction of digital campus,real-time image acquisition equipment is generally installed in classroom.Therefor,there are good conditions to build smart examination rooms.Based on image processing technology,this paper proposes an algorithm to automatically and accurately identify abnormal behaviors in the examination,aiming to maintain the fairness of the examination,improving the efficiency of invigilation and standardizing the exam disciplines further.First,Mosaic technology is used to enhance the video frame image,thereby expanding the image set containing small targets.Secondly,the improved YOLOv5s object detection algorithm is used to locate the human detection box of the candidate’s exam behavior status,quickly and accurately identifying the candidate’s human position.Thirdly,SimplePose is used to locate the key points of the candidate’s body,further accurately describing the posture of the human body and more accurately identifying abnormal behavior.In addition,SimplePose is lightweight and can process the images and extract the key point information quickly to ensure real-time performance.Finally,ConvNeXt image classifier is utilized to classify the human body key point images,which enhances the robustness and stability of the model.The experimental results show that the proposed real-time detection method has a fast detection performance for the identification of cheating behavior of candidates in the examination environment,and the identification accuracy rate reaches 92.1%.It improves the efficiency of examination center supervision and maintains the fairness of the examination.