Home human body tumble detection method based on YOLOv8 improvement

By introducing channel attention mechanism and attitude loss function into the YOLOv8 network, the CA-YOLOv8 model is improved, solving the accuracy and robustness of fall detection in complex environments, achieving higher detection accuracy and lower false alarm rate.

CN119763184AActive Publication Date: 2025-04-04CHONGQING UNIV OF POSTS & TELECOMM
9 Cites 2 Cited by

Patent Information

Application Number
CN202411798257.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-09
Publication Date
2025-04-04
Estimated Expiration
2044-12-09

AI Technical Summary

Technical Problem

In complex environments, existing fall detection methods based on computer vision are difficult to accurately identify human fall behavior, and are easily disturbed by occlusion, light, angle, etc., resulting in a high rate of false alarms and missed alarms.

Method used

Using the improved CA-YOLOv8 network based on YOLOv8, combining the channel attention mechanism and attitude loss function, an improved detection model is established for human fall detection. Through multi-scale feature extraction and feature fusion, this model improves the detection accuracy of fall behavior and enhances the recognition ability of pose changes by introducing pose loss.

Benefits of technology

It improves the accuracy and robustness of fall detection, reduces the probability of false alarms and missed reports, and enhances the model's detection accuracy and adaptability of fall behavior.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to a household human body falling detection method based on YOLOv8 improvement, and belongs to the field of human body posture detection. The method comprises the following steps: firstly, acquiring and labeling a data set, and then inputting the data set into a CA-YOLOv8 model; in the model, image data is subjected to multi-scale feature extraction through a backbone network, and the feature expression ability is enhanced by combining an SE-C2F feature extraction layer with a channel attention mechanism SENet. The extracted features are further fused in the check network, and effective fusion on different scales is ensured. And finally, inputting the fused feature map into a detection head for prediction, and predicting the category and bounding box of the target by adopting a decoupling head. In the training process, attitude loss is introduced, and the attention of the model to the falling target and the recognition capability of attitude change are improved. Through the method, real-time detection of falling of the human body at home can be realized, the detection accuracy and efficiency are improved, and powerful support is provided for home safety.
Need to check novelty before this filing date? Find Prior Art