RBox

RBox stands for rotated box, a geometric primitive used to bound objects with an orientation. It contrasts with axis-aligned bounding boxes (AABB). It is used in computer vision, graphics, robotics, and geographic information systems.

A typical RBox stores the center coordinates (cx, cy), dimensions (w, h), and a rotation angle theta describing orientation. Some libraries store corner points or four-vertex coordinates; some use an angle in degrees or radians, with positive values indicating rotation counterclockwise. In some contexts, it is equivalent to an oriented bounding box (OBB).

RBox can be obtained by rotating an axis-aligned rectangle, by fitting a minimum-area rectangle to a point set (rotating calipers algorithm), or by dataset annotations. Intersection and distance tests are more complex than for AABB due to rotation, and efficient algorithms often rely on the specific parameterization used.

Applications of RBox include object detection and recognition for objects with varying orientation, such as text detection in natural images, aerial imagery analysis, and robotic perception. They are also used in collision detection, layout analysis, and GIS where orientation information improves spatial reasoning.

In software, several libraries provide an RBox or RotatedRect type. OpenCV, for example, uses a RotatedRect representation with center, size, and angle. Other libraries may store an RBox as corner coordinates or as a five-parameter (cx, cy, w, h, angle) representation, depending on the API.