A method for navigation of a small UAV through an unsurveyed environment (e.g. a forest) is presented. In particular the problem of estimating the state of the aircraft and of obstacles in the environment given limited sensors (an inertial measurement unit and a monocular camera) is addressed. The combination of limited observability and the close proximity of the vehicle to the obstacles lead to significant nonlinearities which cause Extended Kaiman Filter (EKF) based approaches to fail. This paper presents an implementation of an Unscented Kalman Filter (UKF) to estimate the locations of obstacles and the state of the UAV based on measurements from the IMU and camera. This solution is applicable to navigation of a 6DOF vehicle in a three dimensional environment and allows generic obstacle avoidance routines to be used. Simulation results are presented for a two-dimensional environment which demonstrate: (a) stability and consistency of the UKF implementation and a comparison with an EKF implementation; (b) fusion with a potential field obstacle avoidance algorithm to enable navigation in an unknown two-dimensional environment.