Dual vision visual fusion improved YOLO-V7 intelligent elevator face recognition model

In response to the current issues of intelligent elevator face recognition models, such as poor real-time performance, low recognition accuracy, weak multi-target detection capabilities, and high missed detection rates, a binocular vision-based improved YOLO-v7 intelligent elevator face recognition model is proposed. Initially, according to the face recognition framework, distortion correction and MATLAB-based camera calibration methods are employed to model the binocular imaging system, ensuring the accuracy of image input. Then, the elevator face recognition model makes use of the YOLO-v7 architecture and uses the Scylla-IoU (SIoU) loss function to measure the actual and forecast bounding box distance precisely. This makes the model more accurate at identifying targets and can handle smaller targets with more ease. The Soft NMS (Non-Maximum Suppression) algorithm is applied to retain detection boxes with lower scores but containing valid targets, reducing the impact of occlusions on target detection and minimizing false positives and missed detections in multi-target recognition tasks. Moreover, the Bottle transformer network is substituted for the original Efficient layer aggregation networks in the model to enhance global attention and feature extraction capabilities, while maintaining computational efficiency and improving recognition accuracy. Finally, extensive experiments using an elevator face dataset are carried out to confirm the rationality and efficiency of the improved model. With Precision, Recall, mAP@0.5, and mAP@0.5–0.9 metrics of 94.18%, 96.23%, 95.18%, and 92.73%, respectively, the experimental findings show that the revised YOLO-v7 model achieves noteworthy performance. The face recognition model exhibits strong feature recognition capabilities, particularly excelling in multi-target detection with a low missed detection rate and high recognition accuracy. The detection time for a single image is 4.1 ms, demonstrating good real-time performance and meeting the requirements for efficient detection, making it suitable for practical application scenarios in intelligent elevators.