Three Dimensions Object Recognition Location Research Based On K-Decision Tree

Computer vision system is a main component of inteiligent robot, which is the basic hard and soft control system. The development of CV accelerate the intelligence of robots , especi- ally the development of 3d picture process system extends the scope of robot application in industry and family. Along with the advancement of Chip-fabrication technology, the 3d camera and algorithm make a big progress in 3d picture process system . Whereas the age of 3d picture process is younger, the 3d process system have a long way to go.In the article, we use a 3d camera to get a full 3d model , and with the starting of 3d object recognition, propose a recognition-mthod based on machine learning technique. We give a complete solution , which improves the method performance in the ways of Feature Engineering and Machine Learning technique.The Feature Engineering of vision objects is a key part of Image recognition,which needs a mount of expert knowledge and experiment experience to find rules of Feature extraction . And the researchers' main work is to find and verify the rules.Our article propose a powerful feature descriptor based on the rules of 3d point cloud feature extraction rules refer to the Point Cloud Library, which is more descrilptive than the other feature in presentation of 3d Poirnt Cloud.There is a specific train and recognition method for a featuire. The main pose recognition methods arte KNN serials and SOM . Now that the SOM approach needs a lot of samples, we select K-Decision Tree belongs to KNT serials as our pose recognition method.In the end ,we make a contrast trial among our feature and VFH ?ESF or other features at. all. The outcome shows that our method gives a higher precision and a low time consuming.