Taking care of individual pigs is important in the management of a group-housed pig farm. However, this is nearly impossible in a large-scale pig farm owing to the shortage of farm workers. Therefore, we propose an automatic monitoring method in this study to solve the management problem of a large-scale pig farm. Particularly, we aim to detect undergrown pigs in group-housed pig rooms by using deep-learning-based computer vision techniques. Because the typical deep learning techniques require a large computational overhead (i.e., Mask-R-CNN), fast and accurate detection of undergrown pigs on an IoT-based embedded device is very challenging. We first obtain the video monitoring data of grouphoused pigs by using a top-view camera that is installed in the pig room, and then detect each moving pig by combining image processing and deep learning techniques. Gaussian Mixture Model is used to detect moving frames and moving objects. In embedded device implementations, by applying deep learning (i.e., TinyYOLO3) only to a few frames with a large number of pixel changes, embedded GPUs can be used efficiently, satisfying the real-time requirement. As a subsequent step, we check the acceptable conditions of the posture and separability from each video frame of the continuous video stream. Finally, to compute the relative size of each pig quickly and accurately, we develop image processing steps to complement the result of deep learning with minimum computational overhead. Furthermore, by pipelining the CPU and GPU steps of a continuous video stream, we can hide the additional image processing time. Based on the experimental results obtained from an embedded device, we confirm that the proposed method can automatically detect undergrown pigs in real-time, by working as an early warning system without any human inspection or measurement of actual weight by a farm worker.

The demand for deep learning frameworks capable of running in edge computing environments is rapidly increasing due to the exponential growth of data volume and the need for real-time processing. However, edge computing environments often have limited resources, necessitating the distribution of deep learning models. Distributing deep learning models can be challenging as it requires specifying the resource type for each process and ensuring that the models are lightweight without performance degradation. To address this issue, we propose the Microservice Deep-learning Edge Detection (MDED) framework, designed for easy deployment and distributed processing in edge computing environments. The MDED framework leverages Docker-based containers and Kubernetes orchestration to obtain a pedestrian-detection deep learning model with a speed of up to 19 FPS, satisfying the semi-real-time condition. The framework employs an ensemble of high-level feature-specific networks (HFN) and low-level feature-specific networks (LFN) trained on the MOT17Det dataset, achieving an accuracy improvement of up to AP₅₀ and AP_0.18 on MOT20Det data.

Objective: To train a deep learning–based markerless gait analysis model from random initialization, tailored for canine gait analysis, in clinical settings and to validate against 2-D marker–based systems. Methods: This prospective observational study used 408 client-owned dogs of more than 30 breeds, ranging from small to large body types. Dogs walked or trotted a 3.3-m straight walkway and were recorded via a lateral single-camera setup using a smartphone camera (1080p resolution, 30 frames/s). Of these, 374 dogs contributed 20,000 markerless images for model training. For model validation and testing, 34 dogs (20 for the validation set and 14 for the test set) were recorded with physical markers that were directly attached to specific bony landmarks by a veterinarian to generate ground-truth data, and each subset included 10 breeds representing a range of sizes and morphologies. The ViTPose-L model was further trained from random initialization on the canine-specific dataset. Accuracy was evaluated using the Common Objects in Context (COCO)-style mean average precision, normalized keypoint error, and percentage of correct keypoints. Results: The markerless model achieved high accuracy, with a mean average precision of 96.6%, mean normalized keypoint error of 2.29 and percentage of correct keypoints at 0.05 of 92.62%. The model’s performance exceeded previously reported benchmarks on public datasets. Conclusions: A deep learning–based markerless gait analysis system shows promising but early-stage accuracy for canine landmark localization compared with 2-D marker–based data. Clinical Relevance: This system may offer clinical utility for canine gait assessment, but further validation across breeds and environments is needed before routine use.