Tof depth data optimization method and device based on unsupervised data

Abstract

The invention discloses a TOF depth data optimization method and device based on unsupervised data, wherein the method includes: acquiring a noise-free human body three-dimensional model database; determining an independent variable for simulating noise; adding noise in the longitudinal direction and the lateral direction respectively ; Render the original 3D model and the 3D model after adding noise, and build an encoder-decoder network; design an energy function and a supervision network, input the network output results into the supervision network, and constrain the feature graphs extracted by the supervision network by convolutional layers. Iterative regression optimization is performed on the parameter weights of the generator and discriminator in the generative adversarial neural network by using the 3D model database of the human body and the energy function until the weights converge. A detailed and noise-free human mesh model. The method uses unsupervised generative adversarial networks on 3D scan data to improve depth data quality while preserving geometric details.

Description

technical field [0001] The invention relates to the technical field of three-dimensional reconstruction in computer vision, in particular to a method and device for optimizing TOF depth data based on unsupervised data. Background technique [0002] With the continuous development of 3D reconstruction technology in the field of computer vision, the 3D reconstruction technology of human body has become a research hotspot in the field of computer room. Using depth cameras to collect depth images to provide additional depth information for 3D reconstruction technology is currently an important direction in research work. [0003] The currently popular consumer-grade monocular depth camera based on the TOF principle continuously emits light pulses (generally invisible light) to the observed object, and then receives the light pulses reflected from the object, and detects the flight (round-trip) time of the light pulses. to calculate the distance of the measured object from the c...

AI Technical Summary

Problems solved by technology

However, when collecting depth images of people, due to the presence of geometric details such as facial features and clothes folds, the use of low-pass filters will cause the loss of fine geometric information at the same time.

On the other hand, the use of deep neural networks to improve the quality of depth data while retaining geometric details has high hopes. However, due to the lack of real depth images paired with real depth data, it is difficult to construct supervised real data for training optimization.

Images