Detect The Position Of Three-Dimensional Objects Based On Kinect

Since kinect came out, As its low price and strong sense of body function, it has broad prospect in visionfield. But the most applications are still limited to thetwo-dimension (2d) space at present, the application of three-dimension (3d)was step by step towards people’s life and production. In this thesis,3d position of the object was obtained by applying kinect to the line of the industrial production.Fist, this thesis introduces research background and its significance, then the structure and function of the device and depth measuring principle were detailed, and color and depth information of the test object were extracted through c#/c++ program, then the information was fusioned, x, y, z, r, g, b point cloud information of the object were outputed.Second, study and analysis were carried out on the kinect measuring precision, the best location to measure object was found, the different methodof filter denoising methods were comparative analysised. In depth direction, maximum of multiple images’depth was used. In x, y direction, two-dimensional gaussian filter algorithm was used.Third, camera calibration algorithm was studied. Relationship between image coordinate system and world coordinate system was found. And the related application experiments were done, feasibility of the algorithm was validated. Then several typical template matching algorithm were compared, considering stability and reliability of the template matching algorithm and the actual needs, based on the shape of the template matching was selected. The location of the object was found by this algorithm, and the transverse and vertical scaling of the object relative to the template were got. And the related application experiments were used to verify the feasibility of the algorithm. Then the algorithm of actual physical size of kinect pixels was studied, And the test of human height detection was done to verify the reliability of the algorithm.At last, Comprehensive experiment of appling kinect to access 3d position of object were done. Color and depth image of objects were got, depth image was mapped to color image, according to the denoising method, the depth information was obtained, according to the template matching, measurement object was found, the objects’distance were estimated by appling pixels’actual size, template matching result was verified by compare it with the actual distance. Last, through the camera calibration algorithm, the location of the objects was found in the world coordinate system. And the result was verified by experiment.