Research And Implemention Of Cattle Face Feature Points Detection

Since cattle facial information is very rich,ranging from skin color,chewing conditions to health status,it is of great importance in cattle disease monitoring.This paper proposes an automatic cattle facial contour scheme on cattle facial extraction to cope with the problems caused by different camera angles,varied illumination and partial occlusion.Three classical human facial contour extraction algorithms are employed in the study: SDM(Supervised Descent Method),LBF(Local Binary Features)and FAAM(Fast Active Appearance Model),which provide a theoretical basis for the further analysis of cattle facial expressions.The main research contents and conclusions are as follows:(1)The study on cattle face detection.Compared with human face,cattle face height-width ratio is much larger.Based on the characteristics of cattle face,this study uses the AdaBoost detector to train cattle face detector.Because cattle face is rather long,we cut the image to the size 15×25.And the background is a negative sample with the same size.Finally,the detection rate of the cattle face detector in 600 cattle face image is 93%.(2)The research on the calibration of cattle face feature points.According to the rules of feature point calibration,29 points are chosen to mark the outline of the cattle face.Manually mark 600 cattle face images,each image feature data saved in text format.All the training sets are aligned,and data sets are aligned to provide data for the subsequent model building.(3)The application of three kinds of contour extraction algorithm in the extraction of cattle face contour.Considering the characteristics of cattle face,we improve the algorithms and optimal our parameters.Also because of the long face,a split model is used to initialize the cattle face model.The first part includes an eye contour with cheeks on both sides,the second part includes the nose and mouth contour.The results show that the accuracy of the contour extraction is improved significantly.We then analyze and compare the time efficiency and the accuracy of the three algorithms.Finally,the performance of each method is evaluated by three different error evaluation criteria and their corresponding computational time costs.The average computational costs of the three contour information extraction methods are 1.48 s,0.39 s and 71.34 s respectively.The average pairwise errors normalized by the facial image size of the three methods are 0.0275 pixel,0.0359 pixel and 0.0269 pixel respectively,and the average pairwise Euclidean errors normalized by the left and right corners of the eyes are 0.0275 pixel,0.0359 pixel and 0.0269 pixel,and the mean square errors are 0.0247 pixel,0.0323 pixel and 0.0242 pixel.The experiment verifies the feasibility and practicability of the facial contour extraction methods.Results show that the FAAM algorithm achieves the highest accuracy with the minimal alignment errors while the LBF algorithm is the most efficient.Therefore,in the process of facial contour extraction,we can choose proper algorithms in varied situations which require different accuracy and efficiency.