Neurotechnologija introduces object-recognition technology

JUNE 27, 2007--SentiSight software-development kit from Neurotechnologija (Vilnius, Lithuania) has been introduced to allow developers to speed integration of computer-based vision systems.

Jun 27th, 2007

JUNE 27, 2007--Humans have a natural ability to learn and recognize objects, even when the objects are presented in different environments, scales, rotations, and poses. There are a growing number of tasks--for example, environmental object recognition for navigation or product recognition and classification in assembly manufacturing--where automating the process of recognizing objects provides an extra measure of safety, efficiency, and cost-effectiveness. For these applications a computer system or robot with object-recognition capabilities is used in conjunction with a camera or other visual input device.

For these reasons, the SentiSight software-development kit (SDK) from Neurotechnologija (Vilnius, Lithuania; www.neurotechnologija.com) has been introduced to allow developers to speed the integration of computer-based vision systems. The SentiSight algorithm provides 2-D and 3-D object recognition for use in robots and machine vision. SentiSight object recognition technology is tolerant to object scale, rotation and can process video streams in real time, enabling its use in autonomous robot navigation, parts identification on an assembly line or road sign recognition in a moving vehicle.

SentiSight SDK enables fully automatic and manual object learning as well as simultaneous multiple object detection and recognition. Using a live camera, series of still images, or video, SentiSight first learns an object by extracting specific features or descriptors of the object from different sides, distances from the camera and angles of view. This enables SentiSight to develop a 2-D or 3-D object model that can be stored in a database. When that same object is later presented to the system, the SentiSight algorithm compares the new images to the existing object model, recognizes the object and outputs the object's name and coordinates.

Because the SentiSight algorithm is tolerant to variations of object scale, rotation, translation and lighting conditions, it can recognize objects in different situations. Results can vary somewhat based on the qualities of the initial learning and the conditions in which the object is to be later recognized.

SentiSight can recognize an object at an average of 10 frames/s for a single object model at 320 x 240 or better resolution. However, for tasks when an even faster response is required, the SentiSight Library includes a tracking mode that enables SentiSight to track an object at speeds of up to 20 frames/s. SentiSight SDK also includes a Camera Manager Library for Microsoft Windows, which allows simultaneous capture from multiple cameras.


More in Cameras & Accessories