Environment Mapping with a Kinect Sensor using Industrial Robots
Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
When integrating an industrial robot to its working cell, an integrator often builds up the scene including the machines surrounding the robot in a virtual environment and performs the programming online. By introducing a precise and correct 3D model of the surroundings of the robot, the integrator must no longer go through the process of building up the environment. Using a low cost Kinect sensor mounted on an industrial robot, a series of 3D scans of the working environmentcan be acquired. By registering the 3D scans, the working environment can be mapped and the integrator can take advantage of the mapping to better adapt the robot to the environment and make the integration more flexible. The resulting 3D model can then be used as a basis for collision free path planning, better recognition and localization of static objects in the scene as workbenches and non-movable machines. This thesis investigates how well the Kinect sensor is suited for building a 3D model of the industrial robot's surroundings, that could be used for industrial robot programming and integration. The point clouds obtained using Kinect were registered together by an initial coarse registration followed by a fine registration. By doing a hand-eye calibration, the pose of Kinect relative to the robot base was obtained, and a transformation for the initial registration could be acquired from the robot itself. The fine registration was done with help of the Iterative Closest Point, which was applied to a set of keypoints extracted from the pair of point clouds to be registered by using the SIFT3D keypoint detector. As the depth data of Kinect is noisy, the data was smoothed out using the Fast Bilateral Filter. The evaluation of this mapping was done both visually and by a comparison to ground-truth data gathered with the industrial robot itself. The results showed that the error of the models obtained varied between negative and positive values, reaching values from approximately -0.5 cm to approximately 1.9 cm. It was concluded that the resulting models could be used for integrating an industrial robot and is suited when a rougher representation of the surroundings is needed where mm precision is not required.
Place, publisher, year, edition, pages
2016. , 66 p.
Engineering and Technology
IdentifiersURN: urn:nbn:se:mdh:diva-30966OAI: oai:DiVA.org:mdh-30966DiVA: diva2:901587