Скачать или смотреть Active Vision-based Control for Fast 3D Reconstruction with an Aerial Robot

In our recent submission to the 2016 International Conference on Robotics and Automation (ICRA), we proposed an online, active control strategy for estimating the three-dimensional structure of objects in a static environment by using a monocular camera mounted on an aerial robot. The pinhole camera model was used as the nonlinear measurement equation for the Extended Kalman Filter, which estimates the unknown depth of each 3D feature with respect to the camera’s current coordinate frame. The robot was actively controlled using an explicit expression for the gradient descent of the estimation error covariance at the next time step. The active technique was then abstracted into a simpler control strategy, which leads to circular trajectories as a function of image feature position. Finally, the two controllers were compared in simulations and in real-time experiments on a quadrotor with a downward facing camera. Our results suggest that moving a camera in a circular trajectory about the object of interest provides near-optimal reconstruction results in a filter-based estimation framework. The active control is computationally efficient and does not require a time consuming optimization in-the-loop, making it best suited for fast, online reconstructions onboard micro aerial vehicles.