Unmanned Aerial Vehicle (UAV) Obstacle Detection Based On Radar And Vision Sensor

Unmanned aerial vehicle(UAV),because of its low cost,good maneuvering performance,cost-effcient,strong survival ability,and no risk of casualties,etc.,has important application value in military,civil and scientific research.However,the existing UAV lacks the ability to detect and avoid obstacles independently.In the unknown complex environment,the vehicle can be controlled by human in the visual range.But beyond visual range,buildings and other obstacles will have a security threat to normal flight.So constructing a set of methods to make the UAV have the ability to perceive and realize the autonomous detection of obstacles is extremely necessary.In this paper,a method for obstacle detection based on millimeter wave radar and vision sensor is proposed.Using millimeter wave radar,the distance and angle of the obstacle can be obtained,which combined with the bottom color of the image,the interest region of the obstacle can be established.Then,the SURF(Speeded Up Robust Feature)algorithm can be used to verify the interest region.The main research contents are as follows:1.Establishing the relationship between radar coordinates and image coordinates.Conventional millimeter wave radar and vision sensor fusion method is usually based on the assumption of two dimensional motion platform,which is very sensitive to the attitude of the platform.Given the change characteristics of height and attitude of UAV platform,the relationship between radar and image coordinate needs to be adaptive to the change of altitude and attitude.Based on the pinhole model of the camera,the relationship between the radar and image coordinate transformation is deduced,which is adaptive to the height and attitude of the UAV.The image coordinates of target obstacles are calculated by fusing the attitude data measured by IMU and the height data measured by differential GPS.2.Obstacle candidate region segmentation.Because the millimeter wave radar can only detect the sparse information of the environment,the full position information of the obstacle cannot be obtained.Therefore,according to the image coordinates of the reflection points of the radar and the image color feature,the candidate region of the obstacle can be segmented.3.Identification of obstacle candidate regions based on SURF algorithm.Firstly,a large number of obstacle images are selected to build the sample database,then the key points of the SURF feature of the sample can be set up.Secondly,by extracting the key points of the SURF features of the candidate regions and comparing the key points of the obstacle sample,the identification and classification of the obstacles are realized.4.Development and experimental verification of the fusion system.Using VS2008 and openCV visual image database,an obstacle detection system is developed,which is integrated with the information of millimeter wave radar,IMU(Inertial Measurement Unit)and GPS(Position System Global).Obstacle detection system is installed in the UAV platform(DIJ S1000+)to carry out the obstacle detection experiments in low altitude environment,and the software is running on the single board computer with a 2.3GHz Celeron processor,a 2G memory.The experimental results show that the proposed method is adaptive to the attitude of the UAV.At the same time,it can quickly and accurately realize on-line detection of obstacles,which turns out to be good real-time performance and accuracy.