Treffer: Depth Image Reconstruction for Enhanced Slam Accuracy in Agricultural Robot Navigation.

This research aims to enhance the accuracy of autonomous positioning for agricultural robots by developing a novel depth image reconstruction method. This method addresses the issue of data loss in depth images, thereby improving the performance of the Simultaneous Localization and Mapping (SLAM) system. An original depth image reconstruction method was developed, which includes: hierarchical multi-scale search for similar blocks and a scale-adaptive priority function, anisotropic gradient computation; and fusion of the found blocks using a neural network architecture consisting of an encoder, a fusion layer, and a decoder. The method was tested on the Rosario dataset, which includes complex agricultural scenarios. The depth image reconstruction demonstrated a significant improvement in quality: the average error (RMSE) decreased by 20-30%, while the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index (SSIM) increased by 20-30% compared to existing methods. It is shown that the proposed method preserves the structure and texture of the restored areas, ensuring accurate reconstruction of large zones with missing pixels. To compare SLAM performance, the S-MSCKF algorithm was selected. The quantitative results for Absolute Trajectory Error (ATE) and the mean RMSE were evaluated using the SLAM system before and after the restoration of the depth maps. The Absolute Trajectory Error (ATE) decreased from 0.62 m to 0.25 m, and the RMSE decreased from 0.85 m to 0.39 m. The new method significantly enhances the accuracy of SLAM systems, especially under challenging conditions such as complex rural landscapes, variable lighting, and longdistance travel. The method has the potential for broad implementation in autonomous control systems for agricultural machinery, increasing the reliability and safety of robot operation. [ABSTRACT FROM AUTHOR]

Copyright of Pakistan Journal of Life & Social Sciences is the property of Elite Scientific Forum and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)