Marker creation and detection of collars

The visual fiducial markers were a core component of this method. Markers were printed and placed on specific cat collars expected to enter the field of view of the video sensors. Using the BeRSTID system, when markers were detected within a field of view, each marker ID’s location, timestamp, and ROI presence were logged. The custom markers used were created using the ArUco module of the OpenCV library as described in OpenCV Detection of ArUco Markers³⁴. Several geometric configurations are supported by ArUco based on the aspect ratio of the 2D surface available and the number of unique subjects. A set of IDs are referred to as a dictionary, and all subjects must have IDs generated from a common dictionary³⁵.

A dictionary of 16 individual 16-bit markers with a square 1:1 aspect ratio was created (Fig. 2A). These were individually printed on collars, with one ID repeatedly displayed sequentially on a single collar (Fig. 2B). To be detected, a marker must be visible to the camera, including a white edge visible surrounding the marker. The computer system will not identify the marker if the white edge is not visible. For specific marker detection methods, see Garrido-Jurado et al.³⁵.

(A) Examples of individual 16-bit ArUco markers of unique markers for individual identification numbers of 0, 1, 2 and 3. (B) TabBand paper veterinary ID collar with a custom sticker, including repeating ID marker. (C) Two cats with unique collar markers. Detection of each ID is visible in annotated output as indicated by “id = [1]” and “id = [2]”, showing individual cat IDs detected in the video.

With the dictionary of fiducial markers generated, a challenge remains in transferring these IDs to the physical media attached to the cat subjects as collars. To achieve this, 25 cm × 2.5 cm custom stickers were repeatedly printed with a single ID across a single sticker. Then, stickers were attached to a standard TabBand paper veterinary identification collar (Fig. 2B). Individual marker IDs on collars allow individual cats with different collars to be detectable in video output (Fig. 2C), and the repeating code allows for multiple opportunities for detection depending on the angle of the cat’s neck.