Multi-sensor patient behavior recognition based on lower limb rehabilitation robot

These instructions give you the basic guidelines for This paper established a multi-sensor fusion system(MSFS) for rehabilitation robots. The system could accurately identify the patient’s behavior state and provide an auxiliary judgment basis for the formulation of rehabilitation robot rehabilitation strategies. It also provided the necessary control input for the underlying controller. The MSFS was based on the mobile lower limb rehabilitation robot (Amber), which was from our laboratory’s research. It collected training data from spatial distribution, kinematics and dynamics. Combined with a secondary behavioural recognition network, it could classify and identify the behavioural state of the user during rehabilitation training. The first stage used a probabilistic neural network(PNN) for real-time determination of whether the patient will fall with 100% accuracy. The second stage network was built by using a support vector machine(SVM) for further detailed classification and identification of the patient’s movement status. We compared the ability of linear kernel function, polynomial kernel function, radial basis kernel function, and Sigmoid kernel function to solve linearly indivisible problems in the original space. Evaluated and analysed through simulation experiments, we found that the SVM-RBF model has the fewest parameters but the highest recognition accuracy. We also compared the accuracy of the BP neural network and the SVM-RBF network in classifying and identifying rehabilitation training behaviour with 92.38% and 97.78% respectively.