Relative Camera Pose Estimation using Synthetic Data with Domain Adaptation via Cycle-Consistent Adversarial Networks

Learning-based visual localization has become prospective over the past decades. Since ground truth pose labels are difficult to obtain, recent methods try to learn pose estimation networks using pixel-perfect synthetic data. However, this also introduces the problem of domain bias. In this paper, we first build a Tuebingen Buildings dataset of RGB images collected by a drone in urban scenes and create a 3D model for each scene. A large number of synthetic images are generated based on these 3D models. We take advantage of image style transfer and cycle-consistent adversarial training to predict the relative camera poses of image pairs based on training over synthetic environment data. We propose a relative camera pose estimation approach to solve the continuous localization problem for autonomous navigation of unmanned systems. Unlike those existing learning-based camera pose estimation methods that train and test in a single scene, our approach successfully estimates the relative camera poses of multiple city locations with a single trained model. We use the Tuebingen Buildings and the Cambridge Landmarks datasets to evaluate the performance of our approach in a single scene and across-scenes. For each dataset, we compare the performance between real images and synthetic images trained models. We also test our model in the indoor dataset 7Scenes to demonstrate its generalization ability.

Modification of MTEA-Based Temperature Drift Error Compensation Model for MEMS-Gyros

The conventional temperature drift error (TDE) compensation model cannot decouple temperature dependence of Si-based materials because temperature correlated quantities (TCQ) have not been obtained comprehensively, and Micro-Electro-Mechanical System gyros’ (MEMS-gyros’) environmental adaptability is reduced in diverse, complicated conditions. The study presents modification of TDE compensation model of MEMS-gyros based on microstructure thermal effect analysis (MTEA). First, Si-based materials’ temperature dependence was studied in microstructure with thermal expansion effect and TCQ that determines the structural deformation were extracted to modify the conventional model, including temperature variation and its square. Second, a precise TDE test method was formed by analyzing heat conduction process between MEMS-gyros and thermal chamber, and temperature experiments were designed and conducted. Third, the modified model’s parameters were identified based on radical basis function artificial neural network (RBF ANN) and its performance was evaluated. Last, the conventional and modified models were compared in performance. The experimental results show MEMS-gyros’ bias stability was up to 10% of the conventional model, the temperature dependence of Si-based materials was decoupled better by the modified one and the environmental adaptability of MEMS-gyros was improved to expand their application in diverse complicated conditions.

The Zurich urban micro aerial vehicle dataset

This paper presents a dataset recorded on-board a camera-equipped micro aerial vehicle flying within the urban streets of Zurich, Switzerland, at low altitudes (i.e. 5–15 m above the ground). The 2 km dataset consists of time synchronized aerial high-resolution images, global position system and inertial measurement unit sensor data, ground-level street view images, and ground truth data. The dataset is ideal to evaluate and benchmark appearance-based localization, monocular visual odometry, simultaneous localization and mapping, and online three-dimensional reconstruction algorithms for micro aerial vehicles in urban environments.

Differential GNSS and Vision-Based Tracking to Improve Navigation Performance in Cooperative Multi-UAV Systems

Autonomous navigation of micro-UAVs is typically based on the integration of low cost Global Navigation Satellite System (GNSS) receivers and Micro-Electro-Mechanical Systems (MEMS)-based inertial and magnetic sensors to stabilize and control the flight. The resulting navigation performance in terms of position and attitude accuracy may not suffice for other mission needs, such as the ones relevant to fine sensor pointing. In this framework, this paper presents a cooperative UAV navigation algorithm that allows a chief vehicle, equipped with inertial and magnetic sensors, a Global Positioning System (GPS) receiver, and a vision system, to improve its navigation performance (in real time or in the post processing phase) exploiting formation flying deputy vehicles equipped with GPS receivers. The focus is set on outdoor environments and the key concept is to exploit differential GPS among vehicles and vision-based tracking (DGPS/Vision) to build a virtual additional navigation sensor whose information is then integrated in a sensor fusion algorithm based on an Extended Kalman Filter. The developed concept and processing architecture are described, with a focus on DGPS/Vision attitude determination algorithm. Performance assessment is carried out on the basis of both numerical simulations and flight tests. In the latter ones, navigation estimates derived from the DGPS/Vision approach are compared with those provided by the onboard autopilot system of a customized quadrotor. The analysis shows the potential of the developed approach, mainly deriving from the possibility to exploit magnetic- and inertial-independent accurate attitude information.