Research And Implementation Of Flight Command Gesture Recognition Based On Deep Learning

With the rapid development of the economy,the throughput of various airports in China continues to increase.In order to improve the efficiency of airport surface operations,this thesis studies an efficient and accurate gesture recognition method for signalman flight command to ensure the safety of airport surface operations.Due to the complexity of airport surface information,the current conventional gesture recognition method based on a visual sensor is difficult to recognize the signalman’s flight command gesture.Artificial feature analyzers are challenging to achieve the human level of accurate recognition of images.Therefore,in this thesis,the convolution attitude machine is used to extract the "bone length" and gravity angle of the skeleton vector as the spatial feature of the flight command gesture,and the short-term memory network is used to extract the temporal feature.This method can achieve a balance between accuracy and speed,enabling the model to recognize flight command gestures in real time.At present,mainstream visual sensor action recognition algorithms are not suitable for recognizing flight command gestures.Firstly,due to the complex and diverse influencing factors involved in real airport scenes command scenarios,such as different weather conditions,changes in light intensity at different time periods,and factors related to the body shape of different signal personnel,it is challenging to extract skeletal features from images containing ground signal personnel.Secondly,it is difficult to reduce or eliminate angle and position interference in the classification process.With the breakthrough of deep learning technology,convolutional pose machines based on computer vision and deep learning have made significant progress in extracting human key points.Therefore,based on computer vision and deep learning,this thesis combines a convolutional attitude machine and a short-term memory network to extract key points and timing features.At the same time,in order to improve the accuracy of flight command gesture recognition,this thesis establishes a flight command gesture video data set(FCGD)with a duration of about 2.5 hours.The data set includes flight command gestures recorded by different signalmen wearing reflective vests in different environments,with changes in lighting,distance,and background complexity,making the verification process more reliable.The experimental results show that the gesture recognition method of flight command based on deep learning proposed in this thesis can effectively recognize the gestures of the ground signalman,and achieve real-time recognition while ensuring accuracy.At the same time,this article also proposes some optimization schemes,such as building a large dataset,optimizing algorithm structure,etc.,to improve detection accuracy and response speed further.The research results of this article have important theoretical and practical significance for the safety of airport surface.