Camlink is a technology probe for exploring tangible computing applications based on visual object recognition. That is, the device allows users to associate certain actions to physical objects, for example, loading and starting a certain playlist of MP3 music files.

We have seen a lot of such applications before, making use of RFID tags or QR codes. The specific feature of Camlink however is, that it does not rely on these technologies. There is no need to attach one of these tags, as Camlink uses machine vision algorithms to recognize objects by their plane visual appearance (no QR code needed, the 'natural' visual pattern of the object is sufficient). The user simply puts an object into the small Camlink box (built for filming the objects under constant light conditions) and the application starts the associated operation. New operations can also be assigned to additional objects and thus the collection of ‘tangible computer commands’ can be extended gradually.

After having tested and evaluated a number of techniques such as Random Forest classifiers, Active Contours, Mean Shift clustering, Watershed Segmentation, the final algorithm of Camlink is quite simple:

• Color space conversion (RGB →HSV)

• Calculation of Euclidean distance between current image (object) and empty box
• Thresholding

• Performing of the Morphological Operation Opening

• Region Labeling / Flood Filling
• Feature extraction and classification by Decision Trees

Camlink was developed using Matlab. The final algorithm was then implemented with C++/openCV. I have evaluated the device's performance and reliability in object recognition, however, it hasn't been explored in an user study with regard to people's needs yet.

Related Images