A Binocular Disparity Calculation Method Based on 3D Convolutional Neural Network

Abstract

The invention relates to a binocular parallax calculation method based on a 3D convolutional neural network. Including: S1. According to the defined multi-scale feature extraction method, perform feature extraction on the input left and right views respectively; S2. Stack the features of the corresponding positions of the corresponding parallax in the left and right images to obtain the 4D cost volume; S3. Use the 3D CNN sub-network for cost aggregation, Obtain the logarithmic likelihood estimation of the disparity value, and upsample to the resolution of the original image, obtain the logarithmic likelihood estimation of the possible disparity value of each pixel, and perform the logarithmic normalization operation to obtain the new logarithmic likelihood Estimate; S4. Calculate the true distribution of the setting; S5. Perform backpropagation training; S6. After obtaining the disparity log likelihood distribution of each pixel, convert it into a probability to obtain the disparity probability distribution; S7. Find the disparity corresponding to the maximum probability Value, S8. Obtain a normalized probability distribution from the aforementioned left and right disparity values ​​and disparity probability distribution; S9. Obtain the final estimated value of each pixel disparity through a weighted average operation. The present invention can effectively improve the accuracy of parallax calculation.

Description

technical field [0001] The invention relates to the field of binocular vision system processing, and more specifically, to a binocular parallax calculation method based on a 3D convolutional neural network. Background technique [0002] As a low-cost method to obtain depth, the binocular vision system has important applications in many fields of robotics. Including map building, obstacle avoidance, positioning, etc. Specifically, it has important applications in the fields of autonomous driving and augmented reality, such as 3D target detection, 3D environment perception, etc. It has the characteristics of low cost, high robustness, strong anti-interference and so on. [0003] Traditional disparity estimation methods usually consist of four parts: feature extraction, cost calculation, cost aggregation, and disparity optimization. With the development of convolutional neural networks and related hardware, CNN estimation of disparity has become a better application. But th...

AI Technical Summary

Problems solved by technology

But there are still many problems, such as the extraction of multi-scale features, high-precision optimization of disparity, etc.

Images