A Study of Feature Combination for Vehicle Detection Based on Image Processing

Figure 1

General scheme of the studied fusion approach. A stream of input images is received from a video captured in a traffic scene, together with the potential locations of vehicles (marked through dashed bounding boxes) obtained via an image segmentation technique. The goal is to verify which of them is actually vehicle. To do so, first individual features (PCA, HOG, etc.) are examined, which provide a support for vehicle existence using an associated classifier. Then, a global decision is made according to a classifier fusion scheme.