Research And Application Of Aircraft Passenger Detection Method Based On Video Sequence

With the continuous development of computer vision technology,automated target detection and statistics have been applied in civil aviation.Pedestrian detection based on deep learning is one of the research hotspots in the field of object detection.Double-checking the actual number of passengers in the cabin after the door is closed is one of the necessary security checks before a flight takes off.The implementation of this security inspection program,due to the limitations of the scene and equipment,is mostly completed by manual method,but the manual method has the problems of high cost,low accuracy and low efficiency.Pedestrian detection methods in cabin crew for testing,can replace the low efficiency of manual inspection way,and can improve the degree of automation and accuracy of inspection process,but traditional pedestrian detection algorithm is lack of accuracy,slow speed,pedestrian detection method based on the deep learning speed can satisfy the requirement of real-time,but low accuracy.In order to solve the above problems,this paper studies pedestrian detection methods based on deep learning and proposes CC-CSP pedestrian detection algorithm on the basis of CSP algorithm.The main work of this paper is as follows:1.To solve the problem of insufficient semantic information extraction by CSP algorithm,this paper proposes a multi-scale CSPN module,which can extract more semantic information by deepening the network depth.At the same time,in order to reduce the burden of computation and memory caused by the increase of network depth,the CSPN structure is adopted in this module,and the maximum semantic information can be obtained with the minimum computational cost through the repeated use of gradient flow information.2.To solve the problem that CSP algorithm fails to effectively fuse shallow features and deep features,this paper proposes a dual attention module based on feature fusion,which can effectively fuse shallow features and deep features from multiple scales.At the same time,this module can reasonably distribute the weights of the fused features in the two dimensions of space and passage,which effectively improves the detection ability of pedestrians.3.In view of the problem that the detection speed and accuracy are reduced due to the limitations of the camera itself and the influence of external lighting conditions,the image pretreatment is carried out by selecting the down-sampling algorithm,median filtering and Gamma transform algorithm of Gamma=0.5 through comparative analysis of experiments,in order to accelerate the detection speed and improve the detection accuracy.4.In order to apply THE CC-CSP algorithm to the actual scene,this paper designed a passenger detection system,which has an image display interface and can automatically count and display the total number of passengers,which has a certain practical value.Finally,this paper carries out experimental verification of CC-CSP algorithm in public data sets and real scenarios.Compared with the CSP algorithm,the proposed CC-CSP algorithm reduced the log-average missing rate(^-2)by 2.60%in the Heavy Occlusion subset of Caltech Pedestrian Open dataset.For the Reasonable subset,^-2 decreased from 4.54%to 4.44%,and for the ALL subset,^-2 decreased by0.98%.In the hard-to-distinguish samples of actual scenes,the average confidence of CC-CSP algorithm proposed in this paper is 48.28%higher than that of CSP algorithm.The above two experimental results show that the pedestrian detection algorithm proposed in this paper has high accuracy and can meet the practical application requirements.