Small Abandoned Object Detection In Highway Scene Based On Background Separation And Gaussian Mixture Model

In recent years,with the continuous extension of highway mileage,the traffic volume has been rising steadily,and the number of accidents caused by highway spills has increased dramatically.In these spills,most of them are Loose goods on trucks or debris from car failures.These small and hard objects pose a great threat to the safety of highway passengers and passengers.Therefore,the real-time detection of small abandoned objects has become an urgent problem.Different from the motor vehicle,non-motor vehicle,pedestrian and other targets on the highway,the abandoned objects do not have the general features on the image.Therefore,the indirect detection of small target abandoned objects is realized by foreground extraction and noise removal in this paper.Firstly,this paper proposes a background separation based on Gaussian mixture model(BS-GMM)algorithm to extract the foreground image of highway video,improves the background division and pixel type judgment of the original Gaussian mixture model,and proposes the concept of background separation to detect the foreground object,which can adapt to the real-time background environment change while realizing the detection of still object.Then,the image preprocessing and target locating are carried out for the foreground extraction results,and the moving state of the foreground target is analyzed by using the IOU matching tracking method.The interference of a large number of random noise is removed,and the suspected small target projectiles are obtained.Finally,two kinds of common small target noise caused by camera shaking and light shadow are removed,and the shaking suppression algorithm based on frequency statistics and the small target abandoned objects verification method based on edge matching are proposed to screen the suspected small target abandoned objects.Applying the Gaussian mixture model based on background separation to the small abandoned objects detection algorithm in actual highway videos,and achieved good results.The algorithm can accurately detect small abandoned objects such as plastic bags,vehicle debris,etc.The minimum target size can be as low as 200 pixels,and can effectively avoid false alarms caused by camera shake,brightness change,shadow shake and other environmental interference.At the same time,the algorithm is also outstanding in computing efficiency,which can be used for real-time detection of small targets abandoned objects in highway video.