Acquisition 3D Kinematics Of Arm Joint With Kinect

Brain Machine Interfaces (BMIs) Technology can help the disabled restore the lost motor function,through directly reading the motor intents from brain and translating them into the commands of artificial auxiliary equipment.For arm is the most important organ for communication,research on neural decoding between the motion of arm or three-dimensional position of wrist and the brain is still very important.Therefore, a convenient and effective method to acquire the accurate motion information of arm is urgently in need.Comparing with traditional motion capture system, the system based on depth information is more convenient to establish and to use, and has the acceptable precision. In this thesis, we propose two effective method for arm motion capture using depth information:(1) Tracking the color frames which are synchronized with the depth frames, then using the tracking result and the depth to calculate the target joint 3D-position. Additionally, we propose a robust and effective tracking method with fusion of mean-shift and particle filtering, comparing with the traditional particle filtering, we decrease the time consumption to 50%. (2)Transform the problem into a simpler per-pixel classification problem, we divide arm into six parts, every part represent a joint or skeleton. We generate a Random-Forest classifier using the real data, and then predict the category for every depth pixel with the classifier, last we use a weighted centroid method to calculate the 3D position of arm joint. Thus we can acquire 3D kinematics of arm joint in dark environment.Finally, we compare the position result with the position acquired by Motion Analysis Eagle system which based on markers, to prove the effectiveness of our method.