Research On Object Recognition Technology Of Home Service Robot Based On Webots

With the development of science and technology,the continuous improvement of human needs and the ageing of society,the industrialization of service robots is imperative.Although many domestic and foreign scholars have conducted long-term research on this,there are still some distances to be achieved in order for robots to effectively and effectively help humans deal with the many problems that they face in daily life.Object recognition is one of the difficulties.The main reason is that Machine vision-based object recognition requires that the recognition accuracy and recognition speed be satisfied at the same time;The realistic environment faced by the robot is complex and changeable,with many interference factors.And the selection of image features will also have a certain impact on it.This article takes the family environment as the recognition scene of the robot,carries on the research design to the above difficult problem,mainly accomplished the following several aspects of work:(1)The SURF algorithm and SVM classifier are combined to extract and train image features.In this process,the problem of high and indeterminate feature dimensions extracted by the SURF algorithm is proposed.The PCA and Bag-of-visual-word algorithms are used to reduce the dimension and cluster.After extracting the most representative principal component,the characteristic of lacking category information whose feature number is uncertain and the PCA dimension is reduced is clustered by Bagof-visual-word algorithm,which is used to make up for the deficiency of PCA algorithm.At the same time,the feature number extracted by the SURF algorithm is determined to be a fixed value,so as to facilitate the training of features by subsequent SVMs.(2)In view of the influence of the complexity of family scenes on the robot recognition,this paper proposes to use the image of the environment as a sample map and use the image pyramid segmentation to segment the test chart.That is,the test chart is divided into blocks to reduce complexity.The effect of the background.(3)The 2D image extracted from a single perspective will inevitably result in loss of information.In order to further improve the accuracy of robot recognition,this paper proposes to use multi-perspective ideas to extract images from multiple angles of the target object to increase the robot's acquisition of target object information.(4)This article mainly uses Webots robot simulation development platform for experimental testing.On this platform,the configuration of OpenCV,family scenes and robots were modeled,and programming was done using c++.The robot's motion control and target object recognition were realized on this platform.The experimental results show that the design of this paper can effectively reduce the impact of complex environments and occlusion situations on the recognition of target objects,and can ensure certain real-time performance while ensuring the recognition accuracy.Because the modeling of the home environment is done as realistic as possible to restore the real scene,therefore,the research design of this article has a certain degree of feasibility and practicality.