There is a need for reliable and efficient methods for monitoring the activity and social behaviour in cows, in order to optimise management in modern dairy farms. This research presents an embedded system that could track individual cows using Ultra-wideband technology. At the same time, social interactions between individuals around the feeding area are analysed with a computer vision module. Detections of the dairy cows’ negative and positive interactions were performed on foreground video stream using a Long-term Recurrent Convolution Networks model. The sensor fusion system was implemented and tested on seven dairy cows during 45 days in an experimental dairy farm. The system performance was evaluated at the feeding area. The real-time locating system based on Ultra-wideband technology reached an accuracy with mean error 0.39m and standard deviation 0.62 m. The accuracy of detecting the affiliative and agonistic social interactions reached 93.2%. This study demonstrates a potential system for monitoring social interactions between dairy cows.

需要一种可靠且有效的方法来监测奶牛的活动和社会行为，以优化现代奶牛场的管理。这项研究提出了一种嵌入式系统，该系统可以使用超宽带技术跟踪单个母牛。同时，通过计算机视觉模块分析了喂养区域周围个体之间的社交互动。使用长期循环卷积网络模型在前景视频流上检测奶牛的负面和正面互动。传感器融合系统已在实验奶牛场的45天内实施并在7头奶牛上进行了测试。在进料区评估系统性能。基于超宽带技术的实时定位系统达到了精度，平均误差为0。39m，标准差0.62 m。隶属关系和激动性社交互动的检测准确率达到93.2％。这项研究展示了一个监控奶牛之间社交互动的潜在系统。