Unmanned aerial vehicles (UAV) are capable of operating tasks in challenging environments of high risk for humans. However, these tasks usually require UAVs to plan motion in environments where GPS is unavailable. This paper implements a UAV system of motion planning and obstacle avoidance in GPS-denied indoor environments with the  visual-inertial odometry (VIO) approach. Collecting measurements from stereo cameras and IMU, the visual-inertial odometry estimates the states of the UAV by jointly minimizing visual and inertial residuals and solving the nonlinear optimization problem. Then we utilize the minimum snap method to generate feasible trajectories based on the estimated states provided by VIO and an obstacle map of the indoor environment. In order to verify the effectiveness of the proposed system, we perform experiments on a UAV platform and the results show that the UAV accurately follows the planned trajectory and avoids obstacles without the aid of GPS.