Feature Extraction on Highways under Time-Varying Illumination Conditions

Video-based surveillance systems have been introduced in highway monitoring as a means of enabling the assessment of traffic conditions by human operators; however, the ongoing development in the aspects of computer imaging hardware and image processing algorithms motivated the interest for the implementation of autonomous highway monitoring and notification systems. An intelligent traffic monitoring implementation demands an efficient video-based front-end, which performs feature extraction in the image domain. The utilized front-end is required to operate flawlessly in the presence of sensor noise and under time varying illumination conditions. The present article analyzes the operation principles of a video-based feature extractor intended for utilization in visual intelligent transportation systems. The proposed implementation performs feature extraction through the application of a block-based statistic segmentation method. Additionally, the algorithm adapts to scene changes. The background adaptation method combines the results of block-based segmentation with a statistic object-based refinement stage which takes into account edge-related metrics.