VI developed a robust and efficient solution to ascertain the presence of an object from a set of multiple objects in any given scenario including cluttered environments.
Our current system can recognise and detect one or multiple objects, if present in the camera view, from a large group of different objects with recognisable and distinguishable feature-points. Our system allows the end user to feed training images of the objects that are to be recognised and automatically analyses the image to identify the target object in a given scene.
Weapon Recognition & Detection
Baggage Identification and Alert
Logo Detection based brand promotions
Structured indoor clutter tolerant
Indoor-environment illumination change tolerant
Speed of detection after system initialisation: Video rate (~25 fps)
360 in-plane rotation distortion tolerant
Can cope-up with distortion of ±150 out-of-plane rotation if trained only with a single view of the object; can be made ±450 out-of-plane distortion tolerant by training the system with multiple views of the same object
Can handle camera-feed of any resolution which is more than 128 x 128 pixels. Note that it has been observed that for high resolution images (size more than 1024px x 1024px pixels), the speed of the system falls slightly below video-rate; Also note that satisfactory results cannot be guaranteed if the camera-feed resolution is less than 128px x 128px
Practically scale change tolerant
Capable of detecting both 3D and 2D objects
System memory requirement ~ 600 MB (if trained with one view of the object)
Capable of the detecting the presence of one among 100 objects with different features at 10 fps
Accuracy achieved: 97.3% (based on in house test results)