PSX-A-1 Late-Breaking: A fully automated system for 3D image acquisition, classification and processing for real-time monitoring growth of finishing cattle in feedlots

Abstract

Abstract Computer vision has been proposed as an alternative to monitor livestock growth, for example using 3D cameras to estimate body weight and condition. However, implementing this technology outdoors and with unrestrained animals poses a great challenge, because of harsh conditions for the equipment, as well as varying lighting and animal movement. Here we propose an automated system to acquire 3D images of individual animals, integrate the animal ID in the image file, and process images. This comprises a fundamental step for implementing a real-time assessment tool of growth of animals grouped in pens, to aid optimal feed management and marketing decision. The study comprised 50 beef cattle of Nellore breed during a finishing phase at DSM Innovation and Applied Center, in Brazil. Animals were tagged with RFID eartags, and 5 Intel RealSense D435 cameras were installed on the top of the water sources in each pen. For each animal approaching the water tank, an RFID antenna read the animal ID and triggered the camera to acquire a top-down view image. Images were then sent to a central computer using Wi-Fi, and later to the cloud. A decision tree algorithm was trained to sort images as usable or unusable, where the retained images had one single animal on the frame, with full body and relatively straight position. Selected frames were then processed using a threshold Otsu’s method to segment the animal body and obtain the necessary biological features, including body length, multiple width and height values, body area and volume. A total of 16,000+ images were collected, from which 3,100 were deemed usable. The algorithm successfully extracted the animal body from each image and obtained the biological features (cross-correlation values above 0.90). The proposed system was demonstrated to be extremely promising and the next stage of the study will be devoted to individual body weight prediction.