Real-time High Resolution Fusion of Depth Maps on GPU

TL;DR

This paper presents a GPU-based system for real-time, high-resolution surface reconstruction using a single moving depth camera, leveraging sparse data structures and advanced algorithms for improved accuracy and performance.

Contribution

It introduces a novel GPU implementation of volumetric fusion using sparse data structures, combined with texture registration and Kalman filtering for enhanced accuracy and robustness.

Findings

Achieved real-time high-resolution surface reconstruction on commodity hardware.

Improved accuracy and robustness through texture registration and Kalman filtering.

Enhanced scanning area and surface detail using sparse TSDF representation.

Abstract

A system for live high quality surface reconstruction using a single moving depth camera on a commodity hardware is presented. High accuracy and real-time frame rate is achieved by utilizing graphics hardware computing capabilities via OpenCL and by using sparse data structure for volumetric surface representation. Depth sensor pose is estimated by combining serial texture registration algorithm with iterative closest points algorithm (ICP) aligning obtained depth map to the estimated scene model. Aligned surface is then fused into the scene. Kalman filter is used to improve fusion quality. Truncated signed distance function (TSDF) stored as block-based sparse buffer is used to represent surface. Use of sparse data structure greatly increases accuracy of scanned surfaces and maximum scanning area. Traditional GPU implementation of volumetric rendering and fusion algorithms were modified to exploit sparsity to achieve desired performance. Incorporation of texture registration for sensor pose estimation and Kalman filter for measurement integration improved accuracy and robustness of scanning process.