CiA^® 462: CANopen device profile for item-detection devices

This profile specifies the CANopen device profile for devices that identify existence, dimension, orientation, or movement of items in their environment, e.g. optical 2D (two-dimensional) or 3D (three-dimensional) camera and laser device. Often these devices are called vision sensors or object-detection devices. An item-detection or object-detection is a vision task for the purpose of detecting, process, and locating semantic items in digital images and videos. The task involves identifying the position and boundaries of items in an image and classifying them into different categories. An item-detection device collects environmental data by some physical effects like light, radar, ultrasonic sound. This data is pixel-oriented and/or matrix-like organized. An item-detection device collects the data once, event-driven, time-based, or stream-oriented. Output of this collection is distance information in some “distance image data”.
 

The "distance image data" is processed in a stage called “item-detection algorithm”. The item detection is optionally controlled by calibration, ROI (regions of interest) definition, configuration data, and possibly by some input at runtime. Output of the item-detection algorithm is a list of items that are detected. Depending on the device, this list is empty, contains only data for one item, or contains data for several items.

An ROI is a region where the item-detection device does its measurements and determined by the vectors V_{op (origin point)}, Vector1, Vector2 and Vector3. Items outside the ROI are not considered.

Some item-detection devices have the ability to detect the depth of a detected item, others just recognizes the front side. This specification offers three modes, “front-side detection” (so-called 2D mode), “front-side detection including distance of detected item to the sensor” (so-called 2,5D mode), as well as “box mode” (so-called 3D mode) where depth detection is enabled.