Design And Research Of A Feeding Manipulator System Based On Facial Posture Human-computer Interaction

A large population in China,including the elderly,disabled,and clinical patients,suffer from hand motor disorders and are unable to eat independently.The loss of independent eating not only causes physical and mental harm to them but also imposes a heavy economic burden on their families.With the increase of the elderly population and the trend of aging in single-person households,low-cost research on assisted feeding systems has become an important sub-topic of China’s elderly care and disability assistance industry.In this regard,this article designs and implements a low-cost meal assistance robotic arm system based on facial posture control for elderly,sick,and disabled people with independent eating disorders.The study focuses on key technologies such as food recognition,multi-object tracking,facial posture recognition,and interaction,and validates the algorithms based on the ROS system.The main content of this thesis is as follows:First,the research background and significance of the meal assistance robot system are introduced,and the current research status of the meal assistance robot system is reviewed.The research status of the key visual algorithms used in this thesis is analyzed,and the content and structure of the thesis are introduced.Second,the system design scheme of the meal assistance robot is elaborated.A low-cost meal assistance robot arm design scheme based on low-cost mechanical components is proposed,including the construction scheme of the electronic control system and the software architecture scheme.The structure of the meal assistance robot arm is designed from aspects such as basic configuration,transmission structure,driving method,and motor selection,and the kinematics,angle position constraints,and activity space of the robotic arm are analyzed.Third,in response to the inadequate food perception of existing disability assistance robot systems,this study conducts integrated application research on real-time food detection and tracking algorithms based on YOLOv5 and Deepsort.A food dataset is autonomously constructed,and the YOLOv5 network is trained using transfer learning.Combined with the Deepsort algorithm,real-time detection and tracking of food are achieved,which can support the visual servoing grasping operation of the robotic arm and overcome the problem of insufficient overall positioning accuracy caused by low-cost mechanical components.Fourth,to solve the difficulty of interaction control in the meal assistance robot system,an application research of facial posture estimation algorithm based on key point detection is carried out.A human-machine interaction key posture discrimination strategy with high real-time and adaptability is designed,and the reliability of the discrimination strategy is improved by using multi-frame voting and statistical filtering algorithms.This can support the smoothness of the human-machine interaction process.Fifth,the system’s interaction control process and feeding process are designed,and joint application experiments are carried out based on the developed robotic arm,integrated algorithms,and designed strategies.The key visual algorithms are integrated and encapsulated under the ROS,and a complete feeding experiment is designed to evaluate the system from aspects such as food grasping success rate,feeding success rate,interaction response time,and successful feeding time.Finally,the work of this thesis is summarized,and the shortcomings of the designed system are analyzed,and improvement directions are pointed out.