Human gesture recognition using a low cost stereo vision in rehab activities

During rehabilitation, patients tend to do several abnormal gestures to indicate their conditions. Since danger might happen to patients, especially without the supervision of therapist, a monitoring system should be developed. In this paper, a preliminary work is conducted to provide an online monitoring system to replace the therapist role to automatically monitor patient during the physical therapy activities by using a stepper. However, the main objective of this paper is to propose methods that can improve recognition rate of human gesture by implying Linear Discriminant Analysis (LDA) on features and then propose Support Vector Machine (SVM) as classifier. In order to accurately identify gesture of patients such as falling down during physical activities, angle features calculated from the information of head and torso positions is proposed as input data. A low cost RGB and depth camera will be used to track and capture the skeleton joint position of the patient. LDA of joint angles is proposed as feature in this research. The feature extracted will be analysed and classified using SVM to recognize the type of gestures performed by the patient during rehabilitation. As any abnormal gesture was recognized, the system will provide information to be used as an alarm for further supervision by the therapist.