Research And Software Design On Object Tracking Algorithm For Airport Ground Surveillance

In recent years,civil aviation industry has developed rapidly,the density of aircraft has increased significantly,which puts forward higher demand for the safety and efficiency of the airport ground.As an important research direction in the field of machine vision,object tracking has broad application prospects in video surveillance,human-computer interaction,automatic driving.The aim of object tracking is to locate the object of interest in the video,and present the annotation in some certain way.However,in practical applications,the performance of object tracking algorithms is always affected by various complex environments,such as weather changes,scale variation,occlusion,and long-term tracking.A robust object tracking algorithm can provide airport administrators with the accurate position and trajectory of aircraft,which will effectively improve the efficiency and safety of airport ground.In summary,this paper does some research on the airport ground object tracking models based on machine vision,realizes the software design and hardware deployment of related algorithms.The specific work content and research results are as follows:1.The single-object tracking algorithm based on SiamRPN has been improved.Considering various problems encountered in the tracking process of airport ground objects,such as weather changes and long-term tracking,global attention module and template update mechanism are designed.The global attention module introduces global modeling information into the feature extraction network through non-local operation,which effectively alleviates the problem of tracking failure due to occlusion.The template update mechanism judges whether the current object template needs to be updated based on the positioning and perceptual loss,and predicts the new template through a meta-learner in the update stage,so that the template can be updated without polluting as much as possible.During the training phase,the robustness of the model to weather changes is improved by means of data augmentation.2.The multi-object tracking algorithm based on YOLOv5s and DeepSORT has been improved.To meet the demand of airport security for the lightweight and convenient deployment of object tracking algorithm,this paper uses the Wasserstein distance as a measure to quantify the object detection model.After quantization,the data type of model parameters is converted from float32 to int8,which reduced the storage usage by 75%.The network structure of DeepSORT model is optimized by the Inception-Res Unit,which improves the feature extraction ability of the network without additional FLOPs.The post-processing of the tracking trajectory is realized by sliding window sampling and normal distribution fitting.3.The software design of the airport ground object tracking platform is realized.In order to improve the practicability of the object tracking algorithm,the operation logic of the model is encapsulated into a visual operation interface,which supports multiple tracking modes,and has functions such as personnel login,registration,and authority management.The data access logic,personnel management and flight information,etc.are designed to facilitate user query and revise based on the MySQL database.4.The hardware deployment of the airport ground object tracking platform is realized.Based on the Jetson Nano artificial intelligence development board,the mobile terminal deployment of the object tracking algorithms is completed.The multi-thread technology is used to realize the concurrent execution of the single object tracking algorithm and optimize the video I/O logic.To make full use of the computing resources of embedded devices,we use TensorRT to accelerate the model.The experimental results show that the inference speed of the optimized model is significantly improved,which can meet the real-time requirements.