Gait Recognition Of Pigs Based On Skeleton Analysis And Gait Energy Image

With the continuous increase in the scale of pig industry in our country,using information technology to improve pig breeding efficiency and health management level has become an inevitable trend in the development of the pig industry.As an important means to improve the level of informatization in breeding industry,computer vision technology can provide automatic monitoring and early warning for abnormal behavior or epidemic situation of pigs.The pig lameness generally predicts the occurrence of some diseases.In order to recognize the normal gait and lame gait of pigs,a gait recognition method of pigs based on skeleton analysis and gait energy image was proposed in this paper.In order to avoid the effect of light,shadow and background on the extraction accuracy of pigs in a complex piggery environment,Microsoft Kinect camera was used to collect the depth images of pigs.Background subtraction method was used to extract the foreground image of the target pig,and the average filter algorithm was used to reduce the influence of noise.Then the optimal threshold is calculated using the OTSU algorithm and the binarization is operated.The final binary image of the target pig is obtained through morphological opening and closing operations.The skeleton analysis method was used to detect the gait cycle of pigs in this paper.Firstly,the pig skeleton was extracted by morphological methods and pruned to obtain a pure pig skeleton.Then endpoints of the pig skeleton were sorted,and the feature vector of the pig skeleton endpoint was constructed through the skeleton path of the adjacent points.Using the similarity between the skeleton endpoint feature vector of the standard pig skeleton and that of the pig skeleton to be measured,the attribution of the pig skeleton endpoints to be measured were determined(determine the specific part of pigs,such as the forelimb endpoint,ear endpoint,etc.).Finally,the gait cycle of pigs was calculated using the quasi-periodic variation of the relative distance of the pig forelimb endpoints.The experimental results show that this method can effectively detect the gait cycle of pigs.The gait energy image is an efficient representation of gait characteristics.The Pig gait energy image(PGEI)was calculated on a gait cycle and two-dimensional principal component analysis(2DPCA)method was used to reduce its dimension.The nearest neighbor classifier was used to recognize normal gait and lame gait of pigs.By testing on the self-built pigs' gait database,when the 18 principal component vectors are taken,the recognition rate reaches 93.25%.The experimental results show that the method can effectively recognize the normal gait and lame gait of pigs.Compared with the PGEI+PCA method,this method has certain advantages in the number of principal components,training time and correct recognition rate.Finally,a blocked PGEI+2DPCA method was proposed for the problem that the feature dimension is too high.By segmenting the PGEI,the area with high contribution to the recognition effect was found out,and experiment was operated in this area.The recognition rate reaches 93%.At the same time,a good recognition effect is achieved,and the feature dimension is also reduced by half.This study provides a new idea for using computer vision technology to recognize the lame gait and abnormal behavior of pigs.