Research On Texture Tactile Reconstruction And Texture Classification Based On Data-drive

In the field of human-computer interaction,with the fast-growing technologies such as VR(Virtual Reality)and AR(Augmented Reality),more and more researchers have been pursuing tactile perception of virtual objects in virtual scenes.As an emerging interaction technology,force-tactile feedback enables users to enjoy their interactions with virtual environments through different forms of force-sense and/or tactile stimuli,enhancing the interactional realism and immersion to be much more natural,lifelike,convenient and reliable.Texture,a basic surface attribute of an object,not only presents the surface features including shape,structure and grain but also molds the tactile perception in human interaction with the real world,thus playing a key role in human perception of the world.To perceive the texture of an object,we can directly touch the surface of the object with our fingers or hold a rigid tool to slide on the surface;in the sliding,vibration tactile feedback that is mediated by the tool,can reflect the collision between the tool and the surface microstructure of the object,thus enabling us to perceive the surface texture.Hence,we can reconstruct and identify object textures by means of vibration tactile perception.In recent years,it is gradually highlighted to model texture tactile perception based on vibration.We can reconstruct and identify the texture of an object by using vibration signals to analyze textural features and using mathematical models to describe them.This technology is not only widely applied in VR and AR but also valuable and applicable to tactile perception assistance for the blind.This paper studied texture tactile modeling algorithms based on vibration information,plus the application of the algorithms in virtual interactive environments,with the content and innovation provided in the following:1)A multi-source data-driven tactile reconstruction algorithm for virtual textures was proposed to enable lifelike tactile perception of virtual objects in a virtual interactive environment.This algorithm,based on the encoder-decoder framework of the Informer model,uses visual and tactile data as the input and extracts features from both texture surface images and user interaction parameters(sliding speed and pressure force)to real-timely generate dynamic vibration acceleration signals;in short,the algorithm integrates textural image features with user interaction features through the encoder-decoder framework,so as to real-timely generate tactile perception signals for different textures in different interaction states.Finally,the paper verified the effectiveness and generalization of the algorithm through the model evaluation experiments in both performance and realism.2)A texture classification algorithm was proposed on the basis of vibration tactile information.This algorithm,focusing on the multi-scale feature and temporal correlation feature of vibration acceleration data,extracts the features of tactile texture information by using the in-depth architecture with multi-scale convolutional neural network(MCNN)and bidirectional long short-term memory(Bi-LSTM)network in parallel,so as to optimize the classification of different textures.3)Based on the two algorithms abovementioned,the paper built a tactile feedback assistance system for virtual textures and also researched the application of virtual texture interaction technology in the education or training of the blind.The blind can perceive different textural features by touching the surfaces of different materials that are simulated by the system,so that they can further understand various materials and perceive the real world more intuitively,thus enjoying their learning and training.Finally,the feasibility of the system was verified through its application experiments.