Stereo Vision-Based Autonomous Grasping Method of Underactuated Robotic Hand
A technology of stereo vision and robotics, applied in the direction of manipulators, program-controlled manipulators, instruments, etc., can solve the problem that complex objects cannot obtain grasping points, etc., achieve good grasping effect, increase the success rate of grasping, and improve operability Effect
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
specific Embodiment approach 1
[0051] The autonomous grasping method of robot underactuated hand based on stereo vision includes the following steps:
[0052] Step 1. For the object to be grasped and its environment, obtain the RGB-D point cloud of the object and the environment through the Kinect sensor, and filter the point cloud;
[0053] Kinect sensor is a 3D vision sensor launched by Microsoft in November 2010. It contains a color camera and a depth camera, which can directly obtain the color map and depth map in the scene, and then generate the point cloud in the scene; however, due to The point cloud generated by Kinect contains the point cloud of all objects in the scene. The number is huge and the features are complex. It takes a lot of machine time to process, which brings trouble to the subsequent burial; Preprocessing, extracting the point cloud of the object in the point cloud, and performing filtering and normal vector extraction to prepare for the subsequent feature extraction;
[0054] Step...
specific Embodiment approach 2
[0098] The specific steps of the process of filtering the point cloud described in step 1 of this embodiment are as follows:
[0099] Step 1.1, use the radius outlier removal filter (RadiusOutlierRemoval filter) to remove outliers;
[0100] A small amount of outliers caused by noise can be removed by using the radius outlier removal filter provided by the PCL library; the filtering process is as follows, assuming that point A is the point that needs to be filtered, first use the Kd_tree search algorithm to count the points A is the center and r is the total number of points inside the sphere. When the number of points is less than the threshold n, it is considered an outlier;
[0101] Step 1.2. Use the average filter to make the surface of the object smoother.
[0102] The influence of white noise can be removed by using the average filter; the filtering process is as follows, assuming that point A is the point that needs to be filtered, first use the Kd_tree search algorithm...
specific Embodiment approach 3
[0104] The specific steps for establishing a grasping planning scheme based on Gaussian process classification described in step 3 of this embodiment are as follows:
[0105] After the above features are obtained, the grasping plan can be obtained through the machine learning method with teachers; the reason for obtaining the grasping plan by using the machine learning method of Gaussian process classifier: 1) The difference between the actual feature and the ideal feature The errors of are generated by noise, so they obey the Gaussian distribution; so these errors can be learned by the Gaussian process; 2) Compared with the support vector machine and the neural network, the Gaussian process classifier is simpler to construct, and only needs to be determined The kernel function and the mean function are enough, and fewer parameters are used at the same time, which makes the parameter optimization simpler, and the parameters are easier to converge; 3) The Gaussian process classi...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More