YOLOv7 and DeepSORT for Intelligent Quail Behavioral Activities Monitoring

The use of modern technology, such as Artificial Intelligence (AI), for livestock farming applications continue to grow. Computer Vision (CV) enables remote, real-time, and noninvasive observation and monitoring of animals by employing detection and tracking algorithms. The integration of Deep Learning (DL) enhances the ability of computers to analyze, interpret and infer the information received. In this study, an Intelligent quail activity monitoring system (I-QAMS) is introduced. It provides detection and monitoring of quail feeding pattern and locomotion activities reared in a cage. In addition, it estimates the activeness based on the level of agility of the poultry animal. These parameters can be used as indicators to assess animal health. YOLOv7, a DL-based single stage detector, is employed for detection and classification of behavioral activities. The DeepSORT algorithm, an algorithm utilizing Kalman filter, Hungarian Algorithm, and Convolutional Neural Network (CCN), is employed for tracking. To evaluate the bird’s movement, the displacement of the centroid is determined using Euclidean distance. A 500-frame video sequence, approximately 20 seconds in length, is utilized for the agility assessment. The YOLOv7 achieved a mAP of 85.28 in training, and an accuracy of 90.49 on test scenes. The results show the potential of CV- and DL-based detectors and trackers such as YOLO and DeepSORT for poultry behavior monitoring and estimation of the animal’s welfare.