Accurate estimation of human motion intention is a hot topic in the area of rehabilitation robot. However, most of existing studies focus on the recognition accuracy, and pay little attention to the safety. In view of
this situation, a safe method for human motion intention recognition is proposed, where K-Nearest Neighbor(KNN) algorithm is used to build the recognition model and the preprocessed sEMG signals are used as
the inputs. Firstly, sEMG signals from seven muscles of human legs and angles of the hip, knee and ankle joints are recorded simultaneously during human walking, from which the training and test datasets can be
built. The whole dataset was divided into ten time series subsets.  Secondly, by comparing the distances between the test samples and the training samples, k nearest neighbors are selected and the associated joint angles can be estimated from calculation of the weight averages of the nearest neighbors. The estimations are filtered to obtain continuous and smooth angle trajectories. Due to that estimations can be maintained
within the ranges of training data, the proposed method can ensure that the estimations are suitable and safe for patient rehabilitation. Finally, validation and comparison experiments were conducted to verify the
method. Three key issues including the number of the nearest neighbors, the weight vector and the distance were considered for the design of a satisfied KNN sEMG-angle model. The root-mean-square errors for hip,
knee and ankle joints are respectively 4:1◦, 5◦, and 3:5◦.