Research On Key Technologies Of Beacon Light Detection

Beacon lights are commonly used as navigational AIDS.By emitting lights of different colors and frequencies,they play a role in terrain warning,wind warning and traffic command.At present,the detection of beacon lights in China is mainly carried out manually.The detection method is not only heavy workload,strict detection requirements,but also time-consuming for a single test.Therefore,it is of great significance to study a real-time detection system for the optical information of beacon lights.Thesis mainly studies the key technologies in the detection process of navigation lights,as follows:（1）Setting up a portable beacon light detection system.The integrated camera head,miniaturized computing platform and built-in power supply module are constructed for the system.At the same time,aiming at the imaging situation of the beacon light in outdoor detection,the optical imaging structure is designed to realize the filtering of the ambient light and the subtracting of the beacon light itself,so as to achieve the accurate detection of the beacon light quality.（2）Using deep learning technology realizing accurate positioning of beacon lights in images.Thesis studies the identification algorithm and tracking algorithm of the beacon lights,also organically combining both,proposing the DT-Net identification and tracking network of the beacon lights.And the lightweight transformation is carried out.Two kinds of lightweight network structures,DT-Net-fire and DT-Net-block,are proposed to realize the fast identification and long-term tracking of the position of the beacon light.（3）Combining deep learning technology and image processing technology,thesis proposes a color detection algorithm for beacon lights,and designs a flashing period and rhythm detection algorithm for beacon light on this basis.The relationship between light intensity and image gray value is studied through experiments.Through the study of optical detection technology of navigation lights,the innovation points and main conclusions of thesis are as follows:（1）The performance of the existing tracking algorithm in the tracking task of the navigation light for a long time: the success rate is 0.52 and the accuracy is 0.34,while the success rate of the DT-Net proposed in this paper is 0.72 and the accuracy is 0.84.Compared with existing tracking algorithms,DT-Net improves the tracking success rate by 38% and accuracy by 147%.The FPS of DT-Net native network is 29.79,and the lightweight DT-Net-block and DT-Net-fire networks are 39.23 and 53.98,respectively,improved by 31.7% and 81.2%.（2）By using traditional image processing technology,the color detection algorithm is realized,and its performance in the color detection task of the navigation light is as follows: the accuracy is 83%.However,the detection accuracy of the small convolutional network constructed in thesis is 97%,which is improved by 16.8%.At the same time,the conversion formula of light intensity to image gray value is fitted by using the experimental results.The final results show that the portable beacon light testing system built in thesis meets the requirements of beacon light testing and has practical application value.