Research On High-precision Fall Detection Algorithm Of Millimeter-wave Radar

Falls can be extremely dangerous for older adults,making timely detection and reporting of falls critical.However,among existing fall detection methods,fall detection for the elderly suffers from both category imbalance and cost sensitivity problems.The category imbalance problem refers to the fact that the number of fall samples in the fall detection dataset is much smaller than the number of normal activity samples.Cost sensitivity is manifested in the fact that the consequences of one missed fall event are far more dangerous than one false alarm.In previous studies,many scholars have ignored the class imbalance problem that exists on the dataset,which has led many classification algorithms to minimize the overall classification error during training by treating all types of errors as equally important,which in turn makes the trained classifier more inclined to the majority class.However,it is crucial to correctly classify an infrequent but important minority class example.The main objective of this research is precisely to propose corresponding solutions to improve the recall and accuracy of fall detection algorithms,starting from cost-sensitive analysis and imbalanced data sets of fall detection algorithms.Therefore,this thesis proposes a high-precision fall detection algorithm based on cost-sensitive convolutional neural network,which constructs a cost-assignment mechanism by setting cost indicators,forms a convolutional neural network model based on cost-sensitive learning,and uses millimeter wave radar to obtain human motion information,and then achieves high-precision detection of fall events by processing and analyzing motion information.The research focuses on the design and implementation of the fall detection algorithm,experimental simulation,and performance evaluation and optimization.First,the advantages of millimeter wave radar in acquiring human motion information are investigated,and the basic principles of fall detection using millimeter wave radar are introduced in detail.Then,the cost-sensitive learning idea is used to construct the cost matrix of the optimal risk function,and the cost-sensitive convolutional neural network model is designed by considering the cost differences caused by different classification errors.Then,the designed model is applied to the fall detection task to achieve effective classification of unbalanced data sets and automatic detection of fall events,which can effectively reduce the misclassification rate and improve the classification accuracy.Finally,the high accuracy and reliability of the proposed algorithm are verified by simulation experiments and practical application cases.The experimental results show that this method reduces the misclassification rate to less than 2% and achieves an average classification accuracy of more than 97.5% when the number of non-falling samples is much larger than the number of falling samples in the test dataset.The recall rate of this method is 97.94%,the precision rate is 97.71%,and the accuracy in identifying confusing actions has been improved by 2.8%.This indicates that the proposed algorithm shows better performance in terms of accuracy and robustness of fall detection,which has important practical value and broad application prospects.