Vehicle recognition for highway lane survey

A prototype computer vision system and framework for the recognition of vehicles from single images is presented. An appearance based pattern matching algorithm which relies on principal component analysis detects and classifies cars and trucks under geometric constraints (front and back views only) independent of scale and location in an image. Since no motion information is required in order to segment vehicles prior to recognition, the proposed system can be used in situations where extensive lane observation time is not available or motion information is unreliable.