Detecting and Tracking Tractor-Trailers Using View-Based Templates

Vehicle tracking and classification is an important application of computer vision. There exist mainly two types of tracking techniques: Appearance-based tracking and Feature-based tracking. Appearance-based techniques generally require a model or template of the target to be tracked. However the model needs to be robust to the multitude of deformations possible in the course of the target’s movement. This research explores the use of a collection of view based templates, instead of a single template, to accurately trace the contours of the tractor-trailers in moving traffic. The tracking begins with creation of a template sequence by manually processing a segment of traffic which contains an exemplar of the desired class of tractortrailers. The processing involves manually segmenting the tractor-trailer from the rest of the scene in each frame. The template sequence is intended to capture the appearance of the vehicle in all possible poses as it moves through the lane. This collection is then used to detect and track similar tractor-trailers during their movement through the lane. Salient features, such as gradient magnitude information, are incorporated for better alignment of the contour during tracking.