Airline Baggage Appearance Transportability Detection Based On Deep Learning

Self-service baggage checked-in must automatically determine the baggage’s appearance transportability for transportation,requiring self-service equipment to have automatic detection and discrimination capabilities.The difficulty of accurate and efficient detection of baggage transportability in complex backgrounds lies in the accurate and real-time detection of baggage,trays,and labels.To this end,a deep learning-based multi-objective tracking algorithm is developed to solve the problem of baggage category,piece count,pallet,and label detection and discrimination.The main work and contributions of the paper include:Firstly,an airline baggage multi-object tracking dataset was created.For the characteristics of complex backgrounds and randomness of passenger’s baggage placement methods in the airport terminal environment,the airport dataset was recorded by overhead angle cameras when passengers use self-service baggage equipment.For the inadequate negative samples of the airport dataset,the check-in dataset that does not satisfy the appearance suitability under various placement forms is recorded in the laboratory simulation environment.Multi-object tracking annotation is performed on all video data according to the need for supervised learning of multi-class multi-objective tracking networks.Secondly,a multi-object tracking network based on sequential hierarchical sampling model is proposed to improve the accuracy of video object detection and tracking.In view of the large-angle movement of the object,the anchor-free detection network is used to provide reliable detection results;given the problems of object motion blur,geometric transformation,and partial occlusion in the process of passengers placing baggage,a sequential hierarchical sampling module is proposed,which guides the network to detect the current difficult object through the object features extracted at different scales at the moment before,and improves the efficiency of detection.Then the distance matrix of the object between frames is calculated by using the object motion features,the detection frame intersection ratio,and the apparent features of the object.The tracking trajectory is obtained by matching the objects with the Hungarian algorithm.Thirdly,in order to obtain the number information of baggage in the appearance transportability,the objects count in the effective area of the conveyor belt is completed according to the object’s tracking trajectory and the double-line counting strategy,and the appearance suitability is calculated by analyzing the number of various objects.Finally,the experimental results show that the proposed multi-object tracking algorithm based on sequential hierarchical sampling layers can meet the task of inspecting baggage’s appearance and transportability in the self-checking environment.