Authors

Abstract

In view of the two problems existing in the tracing of yellow feather broilers in the flat breeding house: the first is the fast location of yellow feather broilers and the second is the tracking accuracy. In this paper, the foreground detection method based on colour features and YOLOv3 algorithm is used to quickly identify yellow feather broilers respectively, and then Kalman filter and Hungarian matching algorithm are used to track yellow feather broilers in the flat breeding house. The traditional algorithm has a poor recognition effect on the aggregation behaviour of broilers, resulting in poor follow-up tracking effect. Through YOLOv3 training and detection, the aggregated broilers can be well separated. The detection precision and recall rate are 98.8% and 87.5% respectively, far exceeding the accuracy and recall rate of the traditional algorithm. The model combining YOLOv3 and tracking algorithm can quickly and accurately identify and track the yellow feather broilers in the flat breeding house, which provides a new method for the detection of the movement rule and motion trail of the yellow feather broilers.

Abstract in Chinese

针对平养舍中黄羽鸡追踪存在的两个问题：一是黄羽鸡的快速定位，二是追踪精确度，本文使用了基于颜色特征的前景检测方法和YOLOv3算法分别对黄羽鸡进行快速地识别，再利用卡尔曼滤波器和匈牙利匹配算法对平养舍中的黄羽鸡进行追踪。传统算法对于鸡的聚集行为的识别效果较差，导致其后续的追踪效果不佳，通过YOLOv3的训练检测，可以很好的将聚集的鸡只分割出来，其检测的精确率和召回率分别为98.8%和87.5%，远超传统算法的精确率和召回率。使用YOLOv3和追踪算法相结合的模型可以快速精确的将平养舍中的黄羽鸡识别出来并进行追踪，这为黄羽鸡的运动规律和活动轨迹检测提供了新方法。