Reserch On Gesture Recognition And Measurement Of Gesture Motion Parameters Based On Dual Cameras

As a research area that has bright prospects and a large research population, artificial intelligence(AI) not only has been applied in all aspect s of daily life and manufacture, but also is constantly generating new products and technology that are rich in creativity and useful functions. Research contents of this thesis mainly concerns the gesture recognition based on dual cameras, measurement of gesture motion parameters and their application in autonomous vehicle control. Therefore, this thesis involves integrate application of machine vision and computer processing.Firstly, in this thesis, images captured by cameras are firstly transformed to HSV color space, which is more sensitive to complexion. Complexion extraction threshold value is then determined by multiple tests, segmentation of skin color region is completed by complexion information, disturbance from small skin color areas, face and neck to gesture is eliminated. After that, the profile is extracted, Graham scan is realized by programming. Convex hull is drawn based on angle determination method. Based on profiles and convex hull, finger root points are located by integrated application of curvature and depth information. In later detection process, Mean Shift algorithm is applied to track the target.Secondly, Hu matrix value of profile point is calculated based on hand profile information. By comparison between 1~10 number gestures mo dule and based on some similarity rules, the matching is carried out and static recognition is fulfilled.Thirdly, before measurement of gesture movement parameters, Zhangzhengyou calibration method is adopted to calibrate inside and outside parameters of cameras. In this way, inside, outside parameters and rotational matrix are obtained. By calculating basic matrix between dual cameras, polar express form of any point is obtained.Finally, 3D coordinates recovery of finger root points and the measurement o f rotation angle and translation vector are finished. The fast matching algorithm is raised based on basic matrix of dual cameras in polar constraints. With this algorithm, corresponding pixel coordinates of finger root points in dual cameras are obtained. With least squard method, 3D coordinate recovery overdetermined equations are solved and space coordinates of finger roots are obtained. This thesis proposes a control strategy that meets three conditions below:(1) the start time must be greater than end time;(2) the gesture must have movement between start and end time;(3) each time after completion of detection, tests are not repeated. In this strategy, invalid small movements can be eliminated effectively, measurement time chaos can be avoided and both cameras can be coordinated to work in synchronization, following time orders. Meanwhile, considering the disability of cameras to detect complete visual blind spots, calculation methods of rotational angles based on integration of experience and computi ng is raised in this thesis.