Research And Application Of Object Pose Estimation Method Based On RGB Image In Virtual Painting

With the continuous development of virtual reality technology,the means of interaction between virtual reality and the real world have become more convenient and diverse.As the core of virtual reality interaction technology,pose estimation has become an important research direction in this field.Traditional pose estimation methods typically require complex calibration processes,such as using multiple cameras for calibration or requiring special markers for labeling.These methods have certain limitations in practical applications.To address this issue,this article uses deep learning methods for pose estimation.By learning model data,the target object can be identified from RGB images and its pose information can be calculated.The main content of the paper is as follows:(1)In response to the problem of inconsistent distribution between small batches of data and training sets in feature extraction networks,as well as information loss or instability caused by batch normalization processing,this paper proposes a new Residual Convolutional Unit module that combines Res Net network for image feature extraction.This module module can better adapt to complex data distribution compared to convolutional blocks in Res Net,making the network more flexible and efficient.(2)In response to the problem of poor accuracy in the CDPN pose estimation network,the translation information solved by the rotating pose estimation network is combined with the translation estimation network to adjust the weight of the translation information in different networks.Finally,the optimal weight information in the current image is obtained through experiments.And a new RPn P rotation estimation network was proposed.Compared to the iterative RANSAC+P3P algorithm,the RPn P algorithm has better operational efficiency while ensuring the same pose accuracy.(3)Based on the above research content,this article applies pose estimation methods to virtual reality painting systems.Capture the target object through a monocular camera for pose estimation,and synchronously generate corresponding strokes in the system while the object is moving,thereby achieving painting function.This system constructs a passive painting system,providing a new direction for virtual reality interaction technology.