Research On The Algorithm Of Rapid Detection And Measurement Of Airport Runways Based On Vision

The large fixed-wing unmanned fighter jet has a long glide distances and a fast glide speed before takoff.And the detection effect of runway line determines whether the large fixedwing UAV can glide and take off strictly according to the established route and direction before takeoff.The traditional runway line detection algorithm can hardly meet the requirements of runway line detection speed and accuracy for UAVs at the same time,and the existing runway line detection algorithm based on deep learning has improved the detection accuracy,but still faces great challenges in improving the detection speed.In this paper,we focus on both runway line detection speed and detection accuracy,and propose a row selection detection algorithm based on unbalanced column sampling.The main research work and contributions of this paper are as follows:(1)In order to optimize and balance the runway line detection speed and detection accuracy,this paper designs a hybrid attention module PACA(Parallel Atrous Convolution Attention),which connects the channel attention module with the spatial attention module in parallel form.A global standard deviation pooling is added to the channel attention module in parallel with the global mean pooling,while an atrous convolution is introduced on the spatial attention module to obtain global features as well as local features on space by multi-scale fusion to cope with variable external environments such as low light and strong light,and to make up for the shortcomings of traditional algorithms and the general convolutional neural networks in this regard.(2)In order to reduce the deviation of the detected pixel coordinates from the actual position,this paper proposes a row selection detection algorithm based on unbalanced column sampling..The algorithm first does the row selection detection based on the pooling of row anchor features,then does the unbalanced sampling detection in the column direction,clusters the first detection results,and then extends a certain rectangular space to the left and right with the center of the clusters as the reference,and performs fine-grained re-detection of the pixels in the rectangular space to get the runway pixel location in the region,so that the detected runway pixel location in the region has less deviation from the actual runway.(3)To verify the effectiveness of the hybrid attention module PACA,ablation experiments of the proposed line selection detection algorithm based on unbalanced column sampling are conducted on the CULane dataset in this paper.And a comparative experiment of classical detection algorithms such as SCNN(Spatial CNN)for runway line detection in nine scenarios is completed on the CULane dataset,and the simulation results show that the proposed algorithm has certain advantages over other algorithms in terms of detection speed and accuracy.(4)To verify the practicality of the algorithm model in this paper,the runway line image acquisition and detection task flow of the simulated in-cockpit camera in the taxiing phase of the UAV are designed in this paper,including image capture,image perspective correction,runway line detection,runway line fitting,and measurement of the deviation angle and deviation distance of the UAV taxiing direction from the runway centerline in the world coordinate system.And simulated experiments are conducted in the environment built by Unity3 D,and the experimental results show that the proposed line selection detection algorithm based on unbalanced column sampling achieves an F1 score of 68% on the CULane dataset,which is not only able to equal the F1 scores of SCNN and SEG,but also7 times faster than them in terms of detection speed and frame rate up to 185 fps.