Design And Implementation Of Vision-based Abandoned Object Detection System For The Car

With the vigorous development of the transportation industry,people are no longer just satisfied with the transport function of vehicles,but more concerned about its comfort and convenience under the society background currently.In this process,due to the high incidence,serious consequences,and neglected characteristics of the abandoned object in the car,it has gradually become an unsolvable social illness.Nowadays,the primitive manual way of checking the seats when passengers get off is no longer meeting the demand.Intelligent video surveillance,as the rapid development technology in recent years,has become one of the effective solutions to find it.Existing vision-based abandoned object detection algorithms utilize monitoring equipment and image processing methods to achieve excellent results in most public places.In scenarios such as parking lots and construction sites,it has achieved good results and solved the public problems.However,in the automotive field,its applications and research are still in the primary stage,which also reflects its great development potential.After investigation,it is found that most of the existing abandoned object detection algorithms only have the function of detection.However,for a complete system,obtaining the category of the abandoned object must be indispensable,especially the identification of hazardous items,which saves lives and property.Therefore,the research direction of this project is the detection and classification of the abandoned object in the car.In the research of abandoned object detection,near-infrared images are used to overcome the influence of changes in illumination,and a detection algorithm is proposed for the car based on near-infrared images.The algorithm design is based on the queue of suspicious objects and abandoned objects,and is improved in two aspects: determining the owner and determining the static state of the same object under different moments.The final experimental results verify that the algorithm in this thesis is not only comparable to other methods on the public data set,but it can still achieve good results in the car scene.In the research of abandoned object classification,a class-specific sparsity augment collaborative representation-based classification(CSSA-CRC)is proposed.Firstly,the label information is added to the objective function,and the training samples are divided into groups according to the categories.Then,two kinds of coefficients are obtained by applying L1 and L2 regularization to the groups.Finally,the sparsity augment operation is performed on the two kinds of coefficients to obtain the expression coefficients used for correct classification.The experimental results prove that the method proposed has excellent image classification performance.Finally,an abandoned object detection system used in the car cockpit is designed and implemented,which has the characteristics of easy operation and large expandable space.The UI interface and runtime screenshots of the system are also shown.