METHOD AND SYSTEM FOR DETECTION OF PEDESTRIAN CROSSING USING A METHOD OF LIGHT WEIGHTED RANDOM FOREST CLASSIFICATION BY A SOFT TARGET LEARNING METHOD

摘要

The present invention relates to a pedestrian detection method using a random forest classification method that is light weighted by a soft target learning method. More specifically, the pedestrian detection method includes, as a pedestrian detection method, (1) a pedestrian candidate region using a flow map on an input image. Detecting; (2) construct a Hough Window Map for the pedestrian candidate area detected in step (1), and based on the configured Hough window map, the optimal scaling image ratio and the region of interest in the scaling image ( Estimating a Region Of Interest); (3) extracting a Harr-Like feature and an Oriented Center Symmetric Local Binary Pattern (OCS-LBP) feature with respect to the optimal scaling image ratio estimated in step (2) and the region of interest in the scaling image; (4) determining, based on the Harr-Like feature and the OCS-LBP feature extracted in the step (3), a pedestrian for the region of interest using a random forest classification method that is light weighted by a soft target learning technique; And (5) determining a final pedestrian area by applying a Non-Maximum Suppression algorithm to the ROI determined as a pedestrian in step (4). In addition, the present invention relates to a pedestrian detection system using a random forest classification method which is light weighted by a soft target learning technique, and more specifically, to a pedestrian detection system, (A) a pedestrian using a flow map on an input image. A pedestrian candidate area detection module detecting a candidate area; (B) construct a Hough Window Map with respect to the pedestrian candidate area detected by the pedestrian candidate area detection module, and based on the configured Hough window map, the optimal scaling image ratio and the ROI in the corresponding scaling image A region of interest estimation module for estimating a region of interest; (C) Harr-Like extracting a Harr-Like feature and an Oriented Center Symmetric Local Binary Pattern (OCS-LBP) feature with respect to the optimal scaling image ratio estimated by the ROI estimation module and the ROI in the corresponding scaling image. Feature and OCS-LBP feature extraction module; (D) The Harr-Like feature and the OCS-LBP feature extracted by the Harr-Like feature and the OCS-LBP feature extraction module are applied to the region of interest using a random forest classification method that is light weighted by a soft target learning technique. A pedestrian determination module for determining whether a pedestrian; And (E) a final pedestrian area determination module configured to determine a final pedestrian area by applying a non-maximum suppression algorithm to the ROI determined as a pedestrian in the pedestrian determination module. According to the pedestrian detection method and system using the random forest classification method which is lightened by the soft target learning method proposed by the present invention, in detecting the pedestrian from the camera image input in real time, Pedestrians are detected using a lightweight random forest classification method that can significantly reduce processing time and amount of memory by reducing the number of trees in the random forest while maintaining performance. Can be detected.