Real-Time RGB-D Camera Pose Estimation in Novel Scenes Using a Relocalisation Cascade

Abstract

Camera pose estimation is an important problem in computer vision, with applications as diverse as simultaneous localisation and mapping, virtual/augmented reality and navigation. Common techniques match the current image against keyframes with known poses coming from a tracker, directly regress the pose, or establish correspondences between keypoints in the current image and points in the scene in order to estimate the pose. In recent years, regression forests have become a popular alternative to establish such correspondences. They achieve accurate results, but have traditionally needed to be trained offline on the target scene, preventing relocalisation in new environments. Recently, we showed how to circumvent this limitation by adapting a pre-trained forest to a new scene on the fly. The adapted forests achieved relocalisation performance that was on par with that of offline forests, and our approach was able to estimate the camera pose in close to real time, which made it desirable for systems that require online relocalisation. In this paper, we present an extension of this work that achieves significantly better relocalisation performance whilst running fully in real time. To achieve this, we make several changes to the original approach: (i) instead of simply accepting the camera pose hypothesis produced by RANSAC without question, we make it possible to score the final few hypotheses it considers using a geometric approach and select the most promising one; (ii) we chain several instantiations of our relocaliser (with different parameter settings) together in a cascade, allowing us to try faster but less accurate relocalisation first, only falling back to slower, more accurate relocalisation as necessary; and (iii) we tune the parameters of our cascade, and the individual relocalisers it contains, to achieve effective overall performance. Taken together, these changes allow us to significantly improve upon the performance our original state-of-the-art method was able to achieve on the well-known 7-Scenes and Stanford 4 Scenes benchmarks. As additional contributions, we present a novel way of visualising the internal behaviour of our forests, and use the insights gleaned from this to show how to entirely circumvent the need to pre-train a forest on a generic scene.

Let's Take This Online: Adapting Scene Coordinate Regression Network Predictions for Online RGB-D Camera Relocalisation

“…At online training time (purple and red boxes), we fill the reservoirs with points from the target scene, which we cluster using Really Quick Shift [14]. At test time (purple and blue boxes), we predict a reservoir for each pixel, and use the point clusters the reservoirs contain to generate correspondences that can be passed to a Kabsch-RANSAC camera pose estimation backend [12] to relocalise the camera: see §2. 4.…”

“…One online local regression approach is that of [13,12], which showed how to adapt the regression forests of [63] for online use in real time. Their approach achieves stateof-the-art performance on the popular 7-Scenes [63] and Stanford 4 Scenes [68] indoor datasets, and also performs well on some of the easier outdoor scenes from Cambridge Landmarks [36,34,35].…”

“…Their approach achieves stateof-the-art performance on the popular 7-Scenes [63] and Stanford 4 Scenes [68] indoor datasets, and also performs well on some of the easier outdoor scenes from Cambridge Landmarks [36,34,35]. However, because their forests use hand-crafted features that were designed for indoor use [63], they struggle [12] to work out-of-the-box on harder outdoor scenes. Whilst it might in principle be possible to solve this problem by hand-crafting new features for outdoor use, doing so could be time-consuming and costly.…”

“…Indeed, the broader trend in machine learning has been towards replacing models such as regression forests with neural networks that can learn suitable features, rather than trying to hand-craft them manually. However, replacing the forests used by [13,12] with networks is not straightforward. To achieve online relocalisation, they rely on the way in which their forests predict leaves containing reservoirs of points to adapt forests between scenes, and it is tricky to see how this scheme can be easily transferred to work with local regression networks, which tend to directly predict individual points in the training scene.…”

Real-Time Highly Accurate Dense Depth on a Power Budget Using an FPGA-CPU Hybrid SoC

“…Index Terms-Heterogeneous, FPGA, real-time, stereo, depth Obtaining information about the 3D structure of a scene is important for many computer vision and robotics applications, e.g. 3D scene reconstruction [1]- [3], camera relocalisation [4]- [6], navigation and obstacle avoidance [7]. Often, this information will be obtained in the form of a depth image, and various options for acquiring such images exist.…”