Research On Perceptual Algorithms For Disabled Robots Based On Gesture Recognition And Blind Roads Semantic Segmentation

Relying on modern information technology,the disability-assisted robot can provide more humanized services for special people,and has wide application prospects in rehabilitation nursing and other work.The perception module of the disability-assisted robot becomes the basis for subsequent path planning and other work because it can sense human-computer interaction information and environmental information.However,existing sensing algorithms are susceptible to environmental interference or conflicts between speed and accuracy,which limits the application of disability-assisted robots.In this paper,the perception module of the disability-assisted robot is taken as the research object.With the help of deep learning theory,two networks are designed to recognize gestures and segment blind roads and crosswalks,thereby improving the performance of the disability-assisted robot in sensing interactive information and environmental information.The specific contents are as follows:1.Aiming at the problem of ignoring the background of most public gesture data sets,a new gesture data set was constructed.The images of this data set are collected from the real environment and labeled with Labeling to form a gesture data set with complex background and diverse scenes,which improves the problem of poor network robustness.2.A new type of gesture recognition network is proposed.The network structure consists of three parts.The basic network uses a stacking strategy of multiple small convolution kernels to reduce the amount of parameters,thereby speeding up the processing speed.At the same time,a feature pyramid attention module and a feature recalibration module are built to amplify the hand information useful for recognition and weaken the interference of background features,so as to enhance the recognition accuracy of the network in complex environments.3.A new blind roads and crosswalks semantic segmentation data set is designed to solve the defects of the existing blind roads and crosswalks semantic segmentation data sets are small,or only the target area is highlighted.In this paper,labellmg is used to label the collected image,so that the pixels in the image are associated with one of the background,blind roads and crosswalks,which enhances the generalization of the network.4.A semantic segmentation network for blind roads and crosswalks is proposed.The network uses depthwise separable convolution to construct the basic network in the encoder,which achieves a significant reduction in the amount of parameters required by the network and increases the segmentation speed.Then,the dense atrous spatial pyramid pooling module and context feature fusion module effectively enhance the effectiveness of information fusion at different levels and improve the segmentation accuracy of the network.Experimental results show that the proposed two networks improve the detection performance of gestures,blind roads and crosswalks,respectively,and help to improve the ability of the disability-assisted robot to perceive interactive information and environmental information.