Design And Implementation Of Autonomous Perception And Navigation System With Micro Aerial Vehicles Based On Stereo Vision

Micro aerial vehicles(MAVs)are widely used in aerial photography and mapping,agricultural plant protection and other fields due to their flexibility,portability,and the capabilities of vertical take-off and landing,high-speed cruising.With the development of industrial MAVs technology,the application field of MAVs expands continuously,such as smart express,search and rescue,electric inspection,etc.,which put forward higher requirements for the autonomy of MAV in handling various complex situations autonomously during the execution of the mission.For example,in the case of GPS denied,the MAV can still accurately locate,autonomously avoid obstacles during flight,dynamically motion planning,and independently detect appropriate location for landing.Therefore,it is critical for MAV expended its application field that enable MAV with the capability of autonomic perception and navigation.In this paper,a complete MAV autonomous perception and navigation solution is proposed and implemented.It includes environment perception algorithm based on stereo vision,motion planning algorithm integrating global path search and local trajectory optimization.In the perception part,the SLAM algorithm based on stereo vision and inertial sensing is used to realize accurate MAV state estimation and construct visual obstacle environment map.In the motion planning part,according to the state of the MAV and the obstacle map information,the hybrid state A star algorithm is used to search for the feasible path to the destination,and then the B-spline trajectory optimization algorithm is used to further optimize the path.A safe and smooth trajectory that meeting the dynamic constraints of the MAV is obtained.Finally,the trajectory server guides the MAV to fly along the re-planned trajectory.In order to facilitate the application deployment of the autonomous MAV system,the general Robot Operating System is adopted as the software architecture.In addition,Gazebo simulator and Air Sim simulator based a complete MAV autonomous navigation software simulation platform is built for algorithm testing and system analysis.A multi-sensor fusion state estimation algorithm based on stereo vision combined with IMU and GPS is proposed to solved the multi-scene sensing problem of MAV.The stereo vision odometry is integrated with IMU and GPS information by means of tight coupling and loose coupling.Therefore,the problem of state estimation of multi-scene and cross-scene of the MAV is solved.In this paper,various experiments are designed in public datasets,simulation environments and real-world scenes to analyze the real-time and accuracy of each functional algorithm and the feasibility of the whole MAV autonomous perception and navigation system.Finally,the proposed autonomous perception and navigation system is deployed on a four-rotors aircraft equipped with NVIDIA TX2 onboard computer,MYNT-EYE stereo camera and Pixhawk flight controller.The MAV system is validated by performing obstacle avoidance flights in read world environments.