Research on Face Detection Based on CBAM Module and Improved YOLOv5 Algorithm in Smart Campus Security

Smart campus security can enhance the security and prevention capabilities of the campus, and optimize the management efficiency of the campus. In order to improve the accuracy of face detection in smart campus security, a face detection algorithm was designed from a high overhead perspective in an uncontrolled environment. Firstly, facial images were collected from different angles to enhance the baseline dataset WiderFace. Secondly, the study introduced MobileNetV3, convolutional block based attention mechanism, and bidirectional feature pyramid structure to improve the You Only Look Once version 5 algorithm, forming the final face detection model. The results showed that the maximum accuracy of the face detection model after data augmentation was 98.73%, which was 6.36% higher than before data augmentation. The parameters of the face detection model and the number of frames transmitted per second are 3143524 and 309, respectively, and the average accuracy of the entire class @. 5:95 and the average F1 value are 93.010 and 0.9785, respectively. The research and design of data augmentation methods can improve the detection accuracy of face detection models, and the designed face detection model has good detection accuracy and speed at high overhead angles, which can greatly enhance the effectiveness of smart campus security and provide technical support for face detection for the safety of students, teachers, and other campus personnel.