Detection of vehicles in a video stream using spatial frequency domain features

Abstract

The paper presents a method for detecting vehicles registered by a surveillance camera using their representation in spatial frequency domain. It mimics the operation of an optical processor with a ring-wedge detector (RWD). The values of the detector outputs constitute a time series of feature vectors. Chosen features, which are least sensitive to ambient light changes and strongly indicate the presence of vehicles are used as descriptors of the content of detection fields. When the values of the features surpass defined thresholds the field is regarded as occupied that is contains a vehicle. The choice of features is made by establishing statistical relations of the feature time series and corresponding reference vehicle presence data.