Automation of lameness detection with vision techniques has a high potential to improve the early recognition of lame cows and would have a positive impact on time efficient herd management. In order to get individual information about gait features from cows passing a corridor in row e. g. after leaving the milking parlour an automatic separation of cows in a sequence is necessary.
The presented results are based on video recordings done on farm, where cows walk from the milking parlour after milking through a corridor in row of 10 to 20 animals. To cope with problems such as stopping for a while in front of the camera, overlap of cows, non-uniform time interval between cows, etc. an algorithm for cow separation is proposed based on local image filtering and statistical analysis of binary images frame by frame. Filters to enhance horizontal and vertical edges in an image are utilized for shadow and background reduction. Binarization on filtered images is made by using statistical analysis. The column-based summation of binarized images related to a threshold is used to decide when the next cow in a row is detected after another has passed already. First results show 95% correct cow separation.