leg tracking and lameness detection method for dairy cows based on occlusion environment

Livestock and poultry farming is gradually developing in the direction of scale and automation, accurate and efficient monitoring of individual information of livestock and poultry helps to objectively evaluate the growth and health status of animals, and individual information of livestock and poultry includes behavioral information and health information. The information of cow's lameness behavior indirectly reflects the health status of her legs, and accurate tracking of cow's legs is an important foundation for the realization of lameness detection. However, the existing intelligent detection of cow lameness is often used at the milking aisle of the farm, and there are problems of occlusion, blurring, and deformation of the cow's leg in this viewpoint, and this study proposes a method of tracking and detecting the cow's leg and lameness based on the occlusion environment for this kind of problem. First of all, The video of cows walking was recorded and pre-processed in an actual farming setting. From this, 50 video segments were chosen as samples for the cow leg multi-target tracking dataset. Each segment featured a walking cow with a duration of 8-20 seconds. The dataset included a total of 27 healthy cows and 23 lame cows. For the dataset samples, the training and testing sets were randomly divided in the ratio of 8:2. 40 videos were used for model training, and 10 videos were used for model testing. Secondly, the improved FairMOT target tracking model of cow leg was designed to obtain the cow leg trajectory information. Specifically, the coordinate attention (CA) module and Squeeze-Excitation attention (SE) module were introduced into the FairMOT Iterative Deep Aggregation(IDA) and Hierarchical Deep Aggregation (HDA), respectively, to improve the model's ability to focus on the important features of the cow's leg when it undergoes shape mutation and blurring in terms of spatial and channel directions, and to reduce the interference of the redundant information; Introducing NSA Kalman filtering in the trajectory state updating phase to obtain more accurate motion states of leg targets and improve tracking when leg occlusion occurs by reducing the measurement weights of low-confidence detection targets; Introducing the AFLink linking mechanism in the tracking post-processing stage to improve the tracking performance of the model. Again, based on the changing law of the cow gait cycle, the curve-fitting method of cow movement trajectory based on the occlusion environment is designed. Finally, the individual thresholds are used to classify the phases of the cow's gait movement, and six gait features, namely, stride length, total gait period, support phase gait period, swing phase gait period, support phase share, and swing phase share, are extracted, which are then inputted into a Navie Bayes model to achieve the detection and classification of limping behavior. The experimental results show that the improved model has good localization accuracy and tracking effect, compared with the original FairMOT model, MOTA (Multiple Object Tracking Accuracy) is improved by 3.79 percentage points, and MOTP (Multiple Object Tracking Percision) and IDs (Total Target ID Switching The total number of identity switches) decreased by 0.022 and 35 times, respectively. The improved model outperforms existing mainstream multi-target tracking models. In addition, CowMOT has more accurate target localization effect, which is more suitable for tracking the cow's leg movement pattern; the classification accuracy of lameness detection is 86.67%, and this method was effective in identifying lameness in dairy cattle. The results of this study provide a reference for subsequent intelligent monitoring of cow lameness behavior.