People/vehicle classification by recurrent motion of skeleton features

Object classification is a major application in video surveillance such as automatic vehicle detection and pedestrian detection, which is to monitor thousands of vehicles and people. In this study, an object classification algorithm is proposed to classify the objects into persons and vehicles despite the presence of shadow and partial occlusion in mid-field video using recurrent motion image (RMI) of skeleton features. In this framework, the background subtraction using a Gaussian mixture model is followed by Gabor filter based shadow removal in order to remove the shadow in the image. The star skeletonisation algorithm is performed on the segmented objects to obtain skeleton features. Then the RMI is computed and it is partitioned into two sections such as top and bottom. Based on the signatures derived from the bottom section of the partitioned RMI using skeleton features, the object is classified into people and vehicles.