Clark N. Taylor

Vision-based target localization from a fixed-wing miniature air vehicle

This paper presents a method for localizing a ground-based object when imaged from a small fixed-wing unmanned aerial vehicle (UAV). Using the pixel location of the target in an image, with measurements of UAV position and attitude, and camera pose angles, the target is localized in world coordinates. This paper presents a study of possible error sources and localization sensitivities to each source. The localization method has been implemented and experimental results are presented demonstrating the localization of a target to within 11 m of its known location

Cooperative target-capturing with inaccurate target information

This paper presents a distributed target-centric formation control strategy for multiple unmanned aerial vehicles (UAVs) in the presence of target motion uncertainty. The formation is maintained around a target using a combination of a consensus protocol and a sliding mode control law. Consensus helps in distributing the target information which is available only to a subset of vehicles. Sliding mode control compensates for the uncertainty in the target information. Hence, collectively the combined strategy enforces each of the vehicles to maintain its respective position in the formation. We show that if at least one vehicle in a group has target information with some uncertainty and the corresponding communication graph is connected, then a target-centric formation can be maintained. The performance of the proposed strategy is illustrated through simulations.

Obstacle Avoidance For Unmanned Air Vehicles Using Image Feature Tracking

This paper discusses a computer vision algorithm and a control law for obstacle avoidance for small unmanned air vehicles using a video camera as the primary sensor. Small UAVs are used for low altitude surveillance ∞ights where unknown obstacles can be encountered. Small UAVs can be given the capability to navigate in uncertain environments if obstacles are identifled. This paper presents an obstacle detection methodology using feature tracking in a forward looking, onboard camera. Features are found using the Harris Corner Detector and tracked through multiple video frames which provides three dimensional localization of the salient features. A sparse three dimensional map of features provides a rough estimate of obstacle locations. The features are grouped into potentially problematic areas using agglomerative clustering. The small UAV then employs a sliding mode control law in the autopilot to avoid obstacles.

Graph-Based Observability Analysis of Bearing-Only Cooperative Localization

In this paper, we investigate the nonlinear observability properties of bearing-only cooperative localization. We establish a link between observability and a graph that represents measurements and communication between the robots. It is shown that graph theoretic properties like the connectivity and the existence of a path between two nodes can be used to explain the observability of the system. We obtain the maximum rank of the observability matrix without global information and derive conditions under which the maximum rank can be achieved. Furthermore, we show that for complete observability, all of the nodes in the graph must have a path to at least two different landmarks of known location.

Improving MAV pose estimation using visual information

We present a system to improve the estimation of MAV location and attitude by combining GPS, IMU and visual information in an unscented Kalman filter framework. Feature points are tracked and combined to create a homography matrix which is used as the measurement input to the filter. We present a novel method to transform uncertainty in feature tracking to uncertainty in the homography. Using a system developed with this framework, we present results which show that this method can substantially increase the accuracy of pose estimation, compared to GPS/IMU alone.

Cooperative navigation of MAVs In GPS denied areas

Cooperative missions for Miniature Air Vehicles (MAVs) require accurate position, velocity, and attitude estimates for all MAVs within the group for its successful completion. This paper details a cooperative methodology for MAV navigation in times of Global Positioning System (GPS) outages or in GPS denied areas. In this method, each MAV estimates position, attitude, and velocity of all MAVs in its sensor range, including itself. Each MAV collects the IMU measurements from each of the neighboring MAVs and fuses these measurements with relative range and bearing measurements taken of every MAV in its sensor range. This collected data is then used to estimate navigation states using an Extended Kalman Filter (EKF). Simulation results presented in this paper demonstrate that this Cooperative Navigation System (CNS) can effectively constrain pose estimation drift in the absence of GPS. We also performed the nonlinear observability analysis to support the improved performance of CNS.

Observability-based local path planning and collision avoidance for micro air vehicles using bearing-only measurements

In this paper we detail an observability based path planning algorithm for Small and Miniature Air Vehicles (MAVs) navigating among multiple static obstacles. Bearing only measurements are utilized to estimate the time-to-collision (TTC) and bearing to obstacles using the extended Kalman filter (EKF). For the error covariance matrix computed by the EKF to be bounded, the system should be observable. We perform a nonlinear observability analysis to obtain the necessary conditions for complete observability. We use these conditions to design a path planning algorithm which simultaneously minimizes the uncertainties in state estimation while avoiding collisions with obstacles. Simulation results show that the planning algorithm successfully solves the single and multiple obstacle avoidance problems for MAVs while improving the estimation accuracy.

GPU Acceleration of Real-time Feature Based Algorithms

Feature tracking is one of the most fundamental tasks in computer vision, being used as a preliminary step to many high-level algorithms. In general, however, the number of features tracked (leading to more accurate high-level algorithms) must be balanced against the computational requirements of the feature tracking algorithm. To enable a large number of features to be tracked in real time without degrading the computational performance of high-level computer vision algorithms, we offload the feature tracking algorithm to the the video card (GPU) found in modern personal computers. Using the GPU allows for tracking an order of magnitude more features than a pure software-based algorithm, with minimal increase in CPU usage. We have demonstrated the computational benefits of GPU-based feature tracking within a real-time video stabilization application.

Observability-based local path planning and obstacle avoidance using bearing-only measurements

In this paper we present an observability-based local path planning and obstacle avoidance technique that utilizes an extended Kalman Filter (EKF) to estimate the time-to-collision (TTC) and bearing to obstacles using bearing-only measurements. To ensure that the error covariance matrix computed by an EKF is bounded, the system should be observable. We perform a nonlinear observability analysis to obtain the necessary conditions for complete observability of the system. These conditions are used to explicitly design a path planning algorithm that enhances observability while simultaneously avoiding collisions with obstacles. We analyze the behavior of the path planning algorithm and specially define the environments where the path planning algorithm will guarantee collision-free paths that lead to a goal configuration. Numerical results show the effectiveness of the planning algorithm in solving single and multiple obstacle avoidance problems while improving the estimation accuracy.

Dynamic input consensus using integrators

The consensus or agreement problem enables a team of agents to agree on certain information variables using a low-bandwidth, dynamic, and sparsely-connected graph. How ever, most prior work on agreement protocols has focused on converging to a single, static variable. In this paper, we propose a consensus filter that accepts dynamically changing inputs at each agent. We analyze several properties of this consensus filter, proving the outputs of the filter converge to a low-pass filtered version of the average of the inputs. Disagreement portions of the inputs can be significantly attenuated through judicious selection of filter parameters.