Multi-agent Special Formation Tracking Control And Trajectory Detection

In recent years,the technology of unmanned aerial vehicle(UAV)has developed rapidly and been widely used in various fields.In order to address large-scale and complex tasks as well as obtain extra performance gains,it is necessary to use multi-agent cooperative system to executed.This paper deals with spherical formation tracking control problem of first-order agents in the unknown spatial flowfileds.By generating a novel adaptive neural flow estimate,a robust spherical formation tracking control is presented.Moreover,the vision-based lane line detection problem is also under our consideration,which is essential to formation tracking motion in the practice.To this end,a novel lane line segmentation network is proposed based on deep learning for lane line detection in the front view when the UAV is flying at a low altitude.Meanwhile a lane line detection scheme is designed by combining both convolutional neural network and traditional image processing technology for the scene that the number of lane lines is uncertain.The main research work and innovations are as follows:(1)Spherical formation tracking control of multi-agents in the unknown flowfileds.The problem of spherical formation tracking control of first-order agents in the totally unknown spatial flowfileds is studied.Different from the adaptive estimation method for the unknown flow with time-invariant parameters,a novel robust spherical formation tracking control with an adaptive neural estimate for the unknown flow is constructed based on the neighbors’ information,which achieves the multi-objectives of spherical landing,orbit tracking and formation motion.The uniform ultimate boundedness of the system errors is proven by the Lyapunov method.The effectiveness of the proposed method is verified by numerical simulations.(2)Spatial pooling network for lane line segmentation.The lane line image segmentation in the front view when the UAV is flying at a low altitude is studied.To solve the occlusion problem,a spatial pooling module is proposed using a combination of horizontal and vertical pooling to achieve pooling in the direction across the lane line,which enhances the information fusion ability in oblique directions and improves the lane line detection performance.To further improve the performance,an end-to-end network for lane line segmentation based on the spatial pooling module is proposed.The experimental results on tu Simple lane line dataset verify that the proposed method enhances the ability of information perception and fusion of the network for the long and narrow lane lines,and improves the performance of lane line segmentation.(3)A lane line detection scheme design with uncertain count of the lane lines.The problem of lane line detection is studied to deal with the scene that the number of lane lines is uncertain.Based on the spatial pooling network and traditional digital image processing methods,a comprehensive lane line detection scheme is proposed.The experiments on the GAIAC lane line dataset verify that the proposed method is more stable and faster than traditional clustering algorithms.In addition,the segmented quadratic polynomial lane line fitting algorithm and the landmark amendment algorithm based on image brightness bring a significant improvement in the landmark prediction.