Trevor Gee

A practical comparison between Zhang's and Tsai's calibration approaches

With the rise of affordable processing power and off-the-shelf apparatus supporting 3D imaging, there is a growing need for reliable and fast calibration tools, enabling timely accurate data gathering. When confronted with a choice of camera calibration tools, Zhangs and Tsais are not only the most cited, but also the most widely available solutions. Zhangs calibration is often chosen by default, based on the assumption that it is more accurate. However, it typically involves extensive manual data gathering when compared to the Tsai approach. Here, we demonstrate that there is no significant accuracy gain between Tsais or Zhangs approach in terms of stereo matching, given the variety of readily available 3D devices tested. Further to this, the trade-off between measurement accuracy compared to setup and data acquisition time is decisively in favour of Tsai. This paper also covers a new algorithm for the extraction of points from images of checkboards attached to calibration objects.

Tsai camera calibration enhanced

The rise of Unmanned Aerial Vehicles for surveying and sensing tasks have created new challenges for quick calibration of sensing systems, which we feel is a critical issue. In this context, calibration is performed often and needs to be achieved as quickly as possible. An approach with minimal user-interaction, which preserves sensing accuracy would be ideal. We propose a version of Tsais camera calibration with an improved distortion model and two non-linear optimization phases to calibrate our UAV equipped with a stereo-camera system. We trialled our proposed calibration approach against three known reconstruction pipelines: an un-calibrated pipeline, a pipeline calibrated using Zhangs approach and a pipeline calibrated using Tsais original approach. Our findings indicate that our approach has competitive accuracy, while requiring far less user-interaction than Zhangs approach.

Lidar guided stereo simultaneous localization and mapping (SLAM) for UAV outdoor 3-D scene reconstruction

Lidars can be extremely useful tools for measuring outdoor geometry. However while lidar measurements are championed for their high accuracy their point clouds are individually rather sparse and lack colour information. In this work the sparse nature of lidar point clouds is addressed by merging multiple lidar scans into a single large point cloud. This is done by restricting the lidar motion to a single axis of translation and then using interpolation and iterative refinement to acquire a denser model by combining co-registered sets of point clouds. This newly constructed model is then used to guide a basic stereo SLAM (simultaneous localization and mapping) algorithm in order to produce a final dense coloured point cloud that preserves the accuracy of the original lidar measurements. Our experiments were performed at various locations using a 16 channel “Puck” Velodyne lidar and a stereo acquisition system consisting of a DJI Phantom quadcopter and a synchronized pair of GoPro HERO 3+ black edition cameras. Results of these experiments demonstrate that the produced reconstructions are both ascetically sound and quantitatively consistent with a set of individual measurements taken around the scene.

A Raspberry Pi 2-based stereo camera depth meter

The Raspberry Pi single-board computer is a low cost, light weight system with small power requirements. It is an attractive embedded computer vision solution for many applications, including that of UAVs. Here, we focus on the Raspberry Pi 2 and demonstrate that, with the addition of a multiplexer and two camera modules, it is able to execute a full stereo matching pipeline, making it a suitable depth metering device for UAV usage. Our experimental results demonstrate that the proposed configuration is capable of performing reasonably accurate depth estimation for a system moving at a rate of 1 ms−1 when in good lighting conditions.

Estimating extrinsic parameters between a stereo rig and a multi-layer lidar using plane matching and circle feature extraction

In this work, we investigate the problem of estimating a rigid transform mapping between a calibrated stereo camera rig and a multi-layer lidar. Such a transform may be used to merge data between these 2 systems, addressing the colourless sparse nature of the lidar data and potentially improving depth estimation from the stereo pairs. The proposed approach features a novel planar calibration object with three circular features allowing for the robust acquisition of corresponding features between sensors. A closed-form registration of correspondences is proposed, leading to the derivation of the required transform. The main appeal of the proposed approach is its conceptually simple formulation and the fact that only a single image from each device is required for calibration. Our experiments were performed on real data captured in outdoor and indoor environments and demonstrate good performance with a Velodyne VLP-16 lidar and GOPRO HERO 3+ Stereo rig.

Robust Tracking in Weakly Dynamic Scenes

Estimating the inter-frame motion of a free-moving camera is important for the reconstruction of large 3-D scene from one or more sequences of frames. This work focuses on scenes with a mixture of dynamic and static elements and proposes an approach to improve tracking in existing 3-D reconstruction algorithms, as well as provide a basis for new types of 3-D reconstructions that are able to construct scenes of moving objects. The main strategy adopted in this work is to group feature points within fixed block-size within the image then to prune groups whose motion deviates from the dominant motions established through majority voting. Our experiments show that the proposed approach performs well in several outdoor dynamic scenes, significantly outperforming typical feature-based and direct pose estimation techniques in footage with moving elements.

Evaluating using GoPro cameras and Tsai's calibration for video-based submerged river-bed reconstruction

In the last decade, consumer-grade waterproof digital cameras have been available more readily, allowing a more straight-forward application of underwater stereo photogrammetry, with the goal of obtaining high-quality DEMs. This paper presents our work into 3D river-bed reconstruction, using a GoPro Hero 3 black edition camera based on Tsais calibration. The objective of this paper is to identify the performance of Tsais calibration for underwater applications and provide other researchers with practical operation guidelines in respect to camera to object distance and various flow strengths encountered. We present the DEM results for a ground truth hemisphere model, chosen for its resemblance with packed natural water-worked gravel particles.

A dedicated lightweight binocular stereo system for real-time depth-map generation

This work describes a light weight dedicated system, capable of generating a sequence of depth-maps computed from image streams acquired from a synchronized pair of GoPro HERO 3+ cameras in real-time. The envisioned purpose is to capture depth-maps from mid-sized drones for computer vision applications (e.g. surveillance and management of ecosystems). The implementation is of modular design, consisting of a dedicated camera synchronisation box, fast lookup based rectification system, a block matching based dense correspondence finder that uses dynamic programming, and a simple disparity-to-depth conversion module. The final output is transmitted to a server via WIFIor G4 LTE cellular Internet connection for further processing. The complete pipeline is implemented on an Android tablet. The main novelty is the systems ability to operate on small portable devices while retaining reasonable quality and real-time performance for outdoor applications. Our experimental results in estuary, forestry and dairy farming environment support this claim.

Tsai's calibration applied for close-range hydraulic engineering research

Proprietary high-frequency industrial stereo-vision systems can be used for close range applications in dynamic environments. Those systems incorporate faster computation and real-time processing. Alternatively, Tsais calibration method is well suited for high-frequency applications using non-proprietary stereo-vision systems. In this paper, we study and compare the performance of using low-cost cameras for close range applications in dynamic environments. The camera configurations tested had substantial lens distortion, and we show how the corner detection method can be adjusted to be better suited for our specific application. We show that using Tsais calibration results in accurate representation of close range objects relevant for hydraulic engineering applications. We discuss in detail the effect of camera to target distance and image coverage for the various camera configurations used in this study. Depth resolution, surface accuracy and efficiency of the acquisition process are used as our recommendation criteria for future application of Tsais calibration in hydraulic engineering research.