| The multi-sensor information source improves the ability of UAV to extract information and estimate the position and attitude state,surrounding environment and target,greatly strengthens its high-precision,intelligent and autonomous operation functions,and can effectively perform various tasks such as industrial patrol inspection,environmental detection,public security and military reconnaissance.Therefore,based on the sub project of the national plan "research and development of intelligent geophysical exploration UAV system",taking the UAV with multi-sensor and small wheelbase as the carrier,this paper explores the body position estimation method under multi-sensor information source and the decision-making and screening process when performing different tasks,and performs target detection The task methods of target ranging and target range estimation and location are studied.The main contents of this paper are summarized as follows:（1）Design of UAV trajectory tracking controller based on LQR。Construct the mathematical model of UAV and linearize it,design the linear quadratic optimal controller LQR and discretize it to control and track the expected trajectory of the mission.Through experiments,this method can stabilize the change of attitude angle during trajectory tracking.Based on the classic F450 rack,a test prototype with multi-sensor bearing and autonomous task capability is designed and built.The ROS algorithm framework is deployed,and the performance of the prototype is tested and evaluated to provide hardware support for task execution.（2）Design of task decision-making system based on vector machine tree structure。Define the transformation relationship of each sensor information source in different coordinate systems,design the airborne computer task mode based on finite state vector machine,construct the tree sensor inspection model and integrate AC automata to accelerate decision matching and screening,so as to realize the evaluation and effective screening of task acquisition data point set based on fuzzy logic.（3）Multi sensor data fusion location estimation scheme and Optimization。The types of environmental information point sets obtained by multi sensors and observation errors are different.Optical flow is designed and introduced as position compensation.The location estimation methods of lidar optical flow and binocular vision optical flow based on extended Kalman filter are analyzed.After simulation and comparison,In a certain structured environment,based on laser / optical flow,it has better inclusiveness to the body speed than binocular vision / optical flow.In the bench experiment,after considering and analyzing the balance of calculation resources and accuracy,the cartographer algorithm in laser slam is selected for optimization and error test and evaluation.The mapping is also used to assist in verifying the target range estimation and positioning,so as to provide guarantee for subsequent experiments.（4）Improve the target detection of the backbone network and visual positioning based on trilateral ranging 。 For the two target detection situations that the target sample label is included and not included in the training set,the small sample learning single case target detection method of the improved backbone network is adopted to improve the accuracy,and the yolov4 target detection method of shufflenetv2 is used to improve the detection speed Reduce computing resources.In the target positioning,the visual ranging method of improved target frame is designed to reduce the measurement error.The inter group matching solution is based on the trilateral ranging and positioning principle.The data point set filtered by fuzzy logic reasoning is introduced into the error curvature range to simulate the range of the projection plane position of the target.The prototype verifies that the above algorithm is reasonable and effective. |
PDF Full Text Request