|M.Sc Student||Kushnir Vladimir|
|Subject||Development of an Artificial Vision System For|
Estimation of Cow Breathing Rate
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Raphael Linker|
|Full Thesis text - in Hebrew|
Cows must dissipate considerable heat as a by-product of milk production, and heat stress is known to reduce significantly milk production and quality. Numerous studies have investigated the use of systems based on fans, sprays and/or misting for effectively cooling cows.
Both spraying and misting systems have high operational costs, many dairy farmers perceive spraying and misting systems as not economically sound. One of the reasons for the poor efficiency of the cooling systems is the overly simplistic way these systems are operated. Typical operation is based on measurements of air temperature, with spraying/misting being activated when a certain threshold is reached.
Dairy farm management would be significantly improved if the operation of cooling devices could be based on direct monitoring of the cows' level of stress, which is the focus of the present study.
The best indicator of heat stress appears to be the core body temperature, followed by breathing rate.To date, core body can be recorded continuously using vaginal or rectal thermometers. However, this is suitable only for research purposes and not for practical applications. By contrast, respiration rates are typically estimated from visual observations of flanks movements, which is not suitable for automatic control purposes. The central hypothesis of the present work was that it is possible to estimate breathing rate of a cow from a real-time series of consecutive pictures taken at a high frame-rate, thus enabling real-time detection of heat stress.
A simple color video camera placed above the barn floor was used to record several video clips. The camera was operated via a custom LabView interface written as part of this work. Each video file was decomposed into consecutive individual frames, which were analyzed to determine the width of the cow.
In order to extract the contour of the "cow" object within each image, the so-called snake algorithm was used. Since this algorithm is iterative and requires an initial contour, in order to perform the extraction automatically the image was rotated and centered beforehand so that the cow appeared within a pre-defined location. After extracting the cow's contour, the measurement point was determined and the width of the cow at this point was calculated. The final stage was to calculate the breathing rate of the cow from the width variation with time. This was achieved by performing fast Fourier transform (FFT) on the signal and identifying the FFT coefficient corresponding to the breathing rate.