System and Method for Detecting Target Rectangles in an Image(Deal 307)

This patent is generally related to locating target rectangles from a stream of video images in real time. Disclosed is an efficient technique for analysing data to accurately locate the centers of target rectangles while rejecting non-target rectangles in the field of view with a high degree of confidence. Disclosed is a technique for detecting a target rectangle in an image by receiving an image from a image source, transform the received image into a grayscale edge detection image and transform the grayscale image into a binary image using a threshold value. The system determines horizontal and vertical sub-exlons for each picture element in the binary image and generate sub-exlon arrays from the determined horizontal and vertical sub-exlons. Also disclosed are techniques for scanning the sub-exlon arrays to identify exlons for each of four corner types and identify the target rectangles using the identified exlons by confirming that each of four corner types are connected by horizontal and vertical lines. The technology may be implemented in online gaming, video tracking, object recognition, 3D pose estimation, etc.