Research on Real-time Object Detection Algorithms for Elevator Intelligent Monitoring Scenarios

The object detection of electric motorcycles and passengers in elevator is of great practical significance in reducing elevator accidents and ensuring the safety of passengers. In complex environments with dense targets in elevator, such as passengers and objects obstructing each other, lead to the problems such as false positives, missed detections, and object detection jitter in existing object detection methods. To address the above issues, we first collected and annotated over fifteen thousand high-quality multi scene target images, which helps to improve the generalization and robustness of the model. A deformable convolution is designed to enhance the extraction of target features at different scales. An occlusion perception attention mechanism is designed to enhance the extraction of occluded target features. A jitter smoothing algorithm is designed to solve the problem of target quantity jitter in recognition. Through comparative experiments with Faster R-CNN, Mamba-YOLO, YOLOv3, YOLOv5, YOLOv8, and YOLO11 algorithms, it was verified that the algorithm proposed in this paper can better detect children targets and mutually occluded targets, with precision of 97.1% and recall of 96.1%. The recognition results of the number of targets are stable. Our algorithm can better meet the needs of practical engineering.