The Application Research On Target Recognition And Tracking Based On Image And Motion-Control Robot

With the diversified development of human-computer interaction, using body movements to control the robot, which is simple and flexible, can replace the complicated operation procedure, making the motion-control robot have become a new direction in human-computer interaction when it shows great superiority in medical treatment and blasting, etc. Therefore, human action imitated by robot has become the focus of the study. Kinect provides a research platform for motion-control robot by recognizing and tracking the human skeleton. However, Kinect can not recognize the joint’s rotation from human body, causing inconvenience for body to control robot.Based on Kinect’s research on body rotation, a method of wearing finger cots is proposed in the present study. With the image recognition technology, the finger cots can be recognized and tracked. Meanwhile, the finger cots angle instead of the human arm’s rotation angle will be calculated, in the aid of controlling the human arm’s rotation, to achieve the purpose of imitating human arms.Firstly, SIFT algorithm is thoroughly studied based on the image processing method. Then, moving target recognition and predication are respectively realized by using image feature matching and Kalman predictor, which reduces the matching area. In this way, the SIFT algorithm is optimized in the application of image recognition and the design scheme is given for the recognition of moving target image in the end.Secondly, the image matching method and window centroid algorithm was studied. Image size to be tested is selected as the target template to calculate the matching center by image color tracking. A platform is constructed by a 2-DOF robot’s head fixed with a single camera and it can recognize and track the red ball correctly.Thirdly, based on DTW (Dynamic Time Warping) algorithm and data glove, the gesture recognition method and sensor method are studied. Then, comparative analysis is made between the above methods and the image processing method in recognition effect, instantaneity and stability, etc. Finally, advantages of image processing method, which is based on finger cots, are pointed out according to practical application.Finally, the Kinect-based robot control system is designed and programmed by using Labview software. And the home-made 4-DOF robotic arm is used as a hardware platform to participate in testing and performance verification. The pre-processing image of human body and red finger cots is formed when they are input into Kinect camera, and then the angle information of finger cots will be acquired by recognizing and tracking its image. By adopting the vector method to analyze Kinect’s three dimensional information of human skeleton, human joint angle information will be gained. Then, the collected information of finger cots and joint angle is entered into 4-DOF robotic arm so as to realize robot control. To testify the recognizing and tracking effect of robotic arm aided by finger cots on human arm rotation, a programme, using the free elbow to control robotic arm, is made to work with the robotic arm to catch the paper cup.Experimental results find out that human can control the robot more flexible under the finger recognition assistant.