Research On Key Technologies Of Bird Target Detection In Airport Bird Detection System

In recent years,with the rapid development of the aviation industry,frequent bird strikes in airports have seriously threatened people’s lives and property.Traditional bird repellent methods have problems such as low efficiency,high cost,and short timeliness.Nowadays,bird repellent technology faces many challenges such as the large number of birds in the airport,small size,and complex gathering conditions.This paper mainly conducts research on key technologies of bird target detection in airport bird detection systems.Aiming at the problem of low accuracy of airport bird detection,this paper proposes a Temporal Variation Filter(TVF)algorithm and a Bird Detection Network based on Gaussian Heatmap Perception(BDGHP-Net)a combined secondary airport bird target detection algorithm.First of all,using the time-domain variation characteristics,a Temporal Variation Filter method is designed to perform first-level detection on the moving targets in the captured infrared video sequence of airport birds,and obtain the moving area(that is,the single bird and the space have much overlap.Secondly,the candidate frames are screened according to the area distribution.If the area of the candidate frame is less than a certain threshold multiple of the average area of the flock of birds;the target frame is considered to be a single target frame and directly output as the detection result;otherwise It is considered that the target frame contains multiple flying birds,and they are sent to the BDGHP-Net network for secondary detection;then,this paper designs a bird target detection network based on Gaussian heat map perception,using the center of the bird target presented in the infrared video.The appearance characteristics of bright and gradually darkening as the distance from the center increases.Gaussian template is used to filter the labeled truth-value rectangular area to generate pseudo-true value samples of bird target with Gaussian distribution.The VGG16 network is used as the backbone network in the network structure.Extract the depth features of the bird target,use MaxPooling filter processing to enhance the significant area of the multi-channel,introduce convolution operation to reduce the number of network parameters,effectively improve the network inference speed;use the mean square error loss function to input the candidate frame area The pixel-level bird target prediction is performed on the image of,and finally the Gaussian heat map predicted by the BDGHP-Net network is post-processed by the watershed segmentation algorithm to obtain the connected areas of each bird.To verify the effectiveness of the proposed airport bird target detection algorithm,the detection performance tests were carried out on the Bird＿a and Bird＿b infrared data sets taken.The harmonic averages of this paper on the Bird＿a and Bird＿b datasets reached 88.7%and 92.2%,respectively.Compared with some mainstream target detectors in recent years,the algorithm of this paper is compared with Faster R-CNN on the Bird＿a dataset.The harmonic averages of,Mask R-CNN,RFB,YOLOv3,HSD and YOLOv4 algorithms are 14.6%,10.5%,7.6%,9.7%,4.8%and 2.9%higher respectively;on the Bird＿b data set,the algorithm in this paper is compared with Faster The harmonic mean values of R-CNN,Mask R-CNN,RFB,YOLOv3,HSD and YOLOv4 algorithms are 12.9%,9.4%,6.5%,8.0%,4.3%and 2.1%higher,respectively,which verifies the effectiveness of this method.In terms of running speed,the frame rate of this article has reached 21.1,which can meet real-time requirements.The experimental results show that the secondary bird target detection algorithm proposed in this paper can effectively detect the birds that appear in the low-altitude area of the airport.Combined with the detection algorithm proposed in this subject,a laser bird repelling robot system based on machine vision is designed.The system is mainly composed of a two-degree-of-freedom mirror,a high-frequency laser transmitter,a web camera,a mirror servo controller,and a video sequence processing module.The video sequence processing module includes target detection,target tracking,and servo control algorithm implementation.The airport bird repellent system developed in this paper can realize automatic positioning,tracking and stimulating to drive away birds,and can effectively solve the problem of frequent bird accidents in airports.