Recognition Of Calf Basic Behaviors Based On Video Analysis

Behaviors are the important indicators of health status for calves. Behavior recognition is fundamental to precision dairy farming. Studies showed that behavioral changes were feasible to provide an early indication of disease. The real-time analysis of the changes in behaviors,including lying, standing, walking, running, and jumping, is crucial for disease prevention.However, the traditional methods of observation by human beings are time-consuming,labor-intensive and the observers are easily exhausted. Considering the limitation of contacting device, we studied the method to recognize basic behaviors of calf using video analysis to improve the welfare of calves. Firstly, a looping algorithm based on maximum connected region was proposed for fast detecting calf target under complex environment.Secondly, a real-time model was built to renew the background and detect the calf’s target efficiently and accurately. Thirdly, the position of the centroid, the ratio of the height and width of the target outline, and the difference of the centroid moving curve were extracted as the features of behaviors. Finally, a classifier based on structure similarity of behavior features was designed to recognize basic behaviors of the calf.The methods and results of this research include:(1)In the perspective of ethology, we conducted the study of dairy calves behaviors.The sensory physiological principle, physiological characteristics and the basic behaviors of calves were studied.(2) Considering the light conditions of the natural environment in commercial farm,we proposed a looping algorithm based on maximum connected the region to detect calf target. The results demonstrated that the accuracy of the algorithm is 90.94% which was4.59% higher than the background subtraction method.(3) The efficiency and accuracy of various background modeling methods were analyzed. Based on the analysis, we divided the target area from the background and merged the true background into the background model in real-time, to update the background model efficiently and accurately. The results showed that the corrected detection rate of background modeling algorithm was 98.98% which 8.32% higher than that the background subtractiondivision.(4)The methods of tracking target and extracting the behavioral features of calf target were studied. The features included motion parameters and shape contour feature. And the behavior features were input into classifier to recognize basic behaviors of calves.(5)The classical theory of structure similarity was studied and improved. We applied the principle based on structure similarity to classify behavioral features. Finally, a classifier based on structure similarity of behavioral features was designed to recognize basic behaviors of the calf.