A Research Of Assessment Models For Human Head Injury In Fall Accidents

Falling is one of the main reasons that endanger the safety of the elderly.It is especially dangerous to have head impacts,which can lead to death.In recent years,the study of accidental falls on the limb injuries of the elderly has been highly valued.The evaluation of the severity of the fall can avoid delays in the elderly due to low pain sensitivity and provide useful information in the doctor's first aid for subsequent falls.Protection research provides relevant reference.In this thesis,the head injury assessment model of human falls is producted.Starting with different human postures during the falls,and carrying out head injury from falls severity assessment for different scenes of video and wearable sensing,and introducing human body key point extraction and classification learning for features Research on extraction and severity assessment model of head injury.The work of the thesis is as follows:1.Based on the posture estimation of the human body fall behavior process based on video data,the 2D joint points of the human body fall video are extracted with OPENPOSE through video data preprocessing,and the missing key points are repaired by Bi LSTM+CNN to realize the 3D human pose estimation during the fall.Experiments show that the average error of the Human3.6 dataset is 49.4mm.2.In view of the possible head injury caused by the fall behavior,use the RGB camera to collect the fall video,and conduct the head injury assessment based on the fall posture.Combine common head injury indicators to classify the severity of head injury after falls,combine key points of the human body and body parts,use global context awareness to pay attention to the long-term memory neural network model to learn features,and output the classification results through softmax to construct different postures A model for evaluating the severity of fall injuries has an accuracy rate of 74.17%.3.For some scenes that cannot be collected,a wearable sensor-based human body fall head injury assessment model is established.By analyzing the correlation between the acceleration of different parts of the body and the acceleration of the head during the fall,the sensor is selected to be worn,and the three-axis acceleration data of the different parts of the human body during the fall are collected,and the multi-convolution-long short-term memory neural network algorithm is proposed for human body fall head injury assessment.The accuracy of the model on the public dataset and self-test dataset reached64.97% and 62.58%,respectively.4.Using Eclipse and Android Studio development tools,combined with My SQL database and Spring Boot framework,the fall severity assessment system for human fall process is completed.The system includes human fall detection and alarm,human fall process video data and fall characteristics display,automatic assessment of the degree of head injury based on video data and wearable sensor data,community alarm processing,patient electronic medical record management,doctors-assisted diagnosis and treatment,follow-up injury recovery tracking and other functions.The system can realize fall detection,severity assessment and subsequent recovery process management of the elderly.