Aiming at the problems that the human motion recognition method based on machine vision and sensors is easy to reveal user privacy and inconvenient to carry, this paper proposes a human motion recognition method based on passive RFID and multi-model fusion. By wearing a flexible liquid metal tag on the human body, the effect of different actions on the RSSI signal of the receiver is utilized, The approximate component and detail component of the signal extracted by wavelet transform are used as fusion features to represent human motion, and the four models of KNN, DT, SVM and LR are fused to construct the Blending model with KNN, DT and SVM as primary learners and LR as secondary learners for motion recognition. The experiments show that the accuracy of the Blending model for the five movements of standing, sitting, walking, running and falling is 97.29%, Compared with individual learners, it has better recognition effect and greatly improves the convenience of human motion recognition.
To address the problem that metal reflection, ambient illumination and different defect size affect the defect detection accuracy in metal defect detection, an AINDANE-Faster R-CNN based metal flat detection method under complex illumination conditions is proposed, which firstly adopts a adaptive and integrated neighborhood dependent approach for nonlinear enhancement(AINDANE) to preprocess the defect images and improve the image brightness to highlight the detail features such as color,profile and texture of defects, and then ResNet50 network is utilized as the defect semantic feature extraction network for the Two-stage Faster R-CNN model. In addition, this paper also constructs a dataset of three defects of metal aluminum plate under low light and uneven light conditions, such as scratches, oil stains and pits, and the method achieves a mean average accuracy of 92.01% on the defect dataset. Compared to existing one-stage surface defect detection methods, the algorithm in this paper is optimal.
KEYWORDS: Data modeling, Machine learning, Body temperature, Data acquisition, Heart, Education and training, Temperature metrology, Blood pressure, Feature extraction, Classification systems
AI education uses and continues to use AI and cognitive science technology to try to understand the essence of learning and teaching, thus establishing a system to help students master new skills or understand new concepts. Facing the huge and complex education database, how to use ai technology to help teachers monitor students' classroom behavior and non-classroom behavior in real time without affecting students' daily learning and life, so as to improve the learning efficiency and reduce the effect of failing classes, has attracted the attention and consideration of scholars and experts at home and abroad. This paper proposes a behavior state classification system for college students based on human physiological information. The system uses intelligent bracelets to collect students' physiological information data, conduct large-scale data preprocessing and feature extraction, and construct a multi-classifier model based on combination strategy to realize the classification of college students' learning, entertainment, and sleep. The experimental results show that the recognition accuracy of the system reaches 95.43%.
In order to improve the problem of inaccurate results in non-contact heart rate detection due to a series of movements of the subject such as breathing, blinking, facial expressions and noise generated by changes in ambient light, the signal is processed in advance using normalisation and wavelet denoising, and then an extreme gradient boosting (XGBoost) algorithm based on a Gaussian process (GP)-based Bayesian optimization method is introduced. The GP-XGBoost machine learning model was introduced to estimate the heart rate. The results show that the estimation error of heart rate by the GP-XGBoost model is significantly reduced compared to that obtained by the conventional method, promoting the practical application of contactless heart rate measurement.
The feature points extracted by the FAST algorithm in the ORB algorithm are densely distributed, and the repeated appearance of the feature points leads to a long time-consuming process in the matching process. Aiming at this problem, this paper proposes an improved method. First, the image is partitioned, then the feature points obtained by the FAST detection algorithm in the original algorithm are sorted, and the feature points that are ranked first and meet the conditions are selected for binary description, and finally the homography matrix is added after hamming matching to further improve the matching. Accuracy. The experimental results show that the algorithm improves the time-consuming of the original algorithm from 1.4s to 0.7s.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.