Detection of cow mounting behavior using region geometry and optical flow characteristics

Computer vision technology may improve the accuracy and robustness of estrus detection in dairy cows. In this study, we proposed a method for detecting mounting behavior in dairy cattle using the geometric and optical flow characteristics of identified image regions in videos taken on dairy farms. Videos captured on a farm often have a complex background, which can interfere with target detection. In this study, we used masking technology to remove the unrelated background, converted the RGB color space to HSV color space, and adjusted the summation coefficients of the HSV channels to improve the contrast between the cows and background images. Subsequently, the proposed Background Subtraction with Color and Texture Features (BSCTF) algorithm was used to detect cow regions. Then, to perform inter-frame differential processing on detection regions, the geometric and optical flow characteristics of the regions were extracted, and seven optimized features were used to construct regional feature vectors. Finally, a support vector machine (SVM) classifier was trained to classify the detected regions into mounting regions and non-mounting regions, which allowed the identification of mounting behavior. We obtained accuracy and omission rates of this detection method of 98.3% and 6.4%, respectively. The average recognition accuracy and false positive rates of the SVM classifier were 90.9% and 4.2%, respectively. These results demonstrate that the proposed method is effective for detecting the mounting behavior of dairy cows and is convenient for further judging whether cows are in estrus.