Detection and tracking of chickens in low-light images using YOLO network and Kalman filter

• Published in: Neural computing & applications Vol. 34; no. 24; pp. 21987 - 21997
• Main Authors: Siriani, Allan Lincoln Rodrigues, Kodaira, Vanessa, Mehdizadeh, Saman Abdanan, de Alencar Nääs, Irenilza, de Moura, Daniella Jorge, Pereira, Danilo Florentino
• Format: Journal Article
• Language: English
• Published: London Springer London 01.12.2022; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Centroids; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Image classification; Image processing; Image Processing and Computer Vision; Kalman filters; Luminous intensity; Moving targets; Original Article; Poultry; Probability and Statistics in Computer Science; Tracking; Video; Deep learning; YOLO v4; Convolutional neural network; Precision livestock farming
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-022-07664-w

Continuous monitoring of chickens’ movement on-farm is a challenge. The present study aimed to associate the modified YOLO v4 model with a bird tracking algorithm based on a Kalman filter to identify a chicken’s movement using low-resolution video. The videos were captured in grayscale using a top-view camera with a low resolution of 702 × 480 pixels, preventing the application of usual image processing techniques. We used YOLO to extract the characteristics of the image and classification automatically. A dataset with images of tagged chickens was used to detect chickens, being 1000 frames tagged in different videos. The generated model was applied in a video that returned the bounding box of the location of the chicken in the frame. With the limits of the box, the centroid was calculated and exported in a CSV file for tracking processing. The Kalman filter was implemented to track chickens in low light intensity. Results indicated that YOLO presented a 99.9% accuracy in detecting chickens in low-quality videos. Using the Kalman filter, the algorithm tracks the chickens and gives them a particular identification number until they leave the compartment. Furthermore, each moving chicken is located in different colors along with the maps below the image, making chicken detection more convenient. The tracking results of chickens show that the proposed method can correctly handle the new entry and exit moving targets in crowded conditions.