Accurate Boundary Location from Motion

The ability to monitor a visual scene containing animals and to draw intelligent conclusions automatically would have a significant impact on agricultural practice. For example, if the gait of an animal could be objectively measured, early detection of lameness would be possible. If the motion of a sow and piglets could be analysed, a stockman could be alerted if the piglets were in danger of being crushed or were not feeding properly. This work forms part of a programme to estimate the weight and hence growth rate of animals from images. In this case, accurate boundaries are required. Animals are often found in visual situations where the background is cluttered and cannot easily be controlled. Also their own surface is often marked either naturally or by contamination from their environment. Segmentation techniques based on thresholding are usually not successful but it may be possible to exploit the fact that animals move whereas the background is stationary.