Research On Pose Estimation Algorithm Of Parts In Point Cloud Model Based On Virtual 3D Sensor

In the 3C(Computer,Communication,and Consumer Electronics)industry,the automated production is an inevitable choice to reduce production costs and increase production efficiency,especially at the assembly stage where production costs are currently high.Automated assembly of parts is very important and necessary.There are many space assembly behaviors that require six degrees of freedom adjustments.At present,2D cameras widely used in industrial scenes cannot meet the space assembly requirements.What's more,there are not many researches on the pose estimation of 3C consumer electronics products represented by smart phones.Therefore,the goal is to realize the pose estimation of smart phones based on the 3D visual technology.In this dissertation,the 3D vision sensor-binocular structure light is used to obtain the target point cloud view of the object model.The pose estimation of object model is transformed into the pose estimation of 3D camera and then it is transformed into the pose estimation of 3D virtual camera which can from the target point cloud view.In the particle filter algorithm framework,the pose estimation algorithm is achieved with the combination of the virtual 3D camera point cloud view imaging algorithm and ICP motion model.Particles are sprinkled near the point cloud model of the smart phone.Each particle is equivalent to a virtual 3D camera.Each particle uses the HPR algorithm and coordination transformation to extract the 3D point cloud view in the object point cloud model in different density of point cloud at different sights.In the prediction stage,the ICP algorithm is used to drive all particles toward a particular pose where the particle can form the target point cloud view.The algorithm iteratively calculates until the pose of the virtual 3D camera is found that can form the target point cloud view.And then the pose of the object model is calculated by the inverse of the virtual 3D camera pose.In the experimental scheme of pose error measurement,in order to avoid the introduction of hand-eye calibration error,the pose error measurement experiment is divided into position error measurement experiment and orientation error measurement experiment.The pose searching speed in the six-dimensional pose space is accelerated by implementing the virtual 3D camera point cloud view imaging algorithm.The number of particles in the range of 12 to 20 is used.The position error is less than 1.5 mm and the orientation error is less than the degree of 1.5 after two or three iterations of pose estimation algorithm.