StaticPigDetv2: Performance Improvement of Unseen Pig Monitoring Environment Using Depth-Based Background and Facility Information

초록

Highlights The three graphs present overall performance improvements in accuracy and latency across different YOLO models. The red lines indicate the baseline models, while the blue lines indicate the proposed models. The black lines represent the performance changes for each model, highlighting the improvement from the baseline to the proposed method. What are the main findings? What is the implication of the main finding? With YOLOv8, YOLOv10, and YOLOv12 nano models, the proposed method improves the baseline accuracy, while slightly increasing the execution time due to the fixed preprocessing operations, by exploiting the background and facility information. With YOLOv8, YOLOv10, and YOLOv12 small, medium, and large models, however, the proposed method improves both the baseline accuracy and execution time because the portion of the fixed preprocessing operations is relatively decreased with bigger models.Highlights The three graphs present overall performance improvements in accuracy and latency across different YOLO models. The red lines indicate the baseline models, while the blue lines indicate the proposed models. The black lines represent the performance changes for each model, highlighting the improvement from the baseline to the proposed method. What are the main findings? What is the implication of the main finding? With YOLOv8, YOLOv10, and YOLOv12 nano models, the proposed method improves the baseline accuracy, while slightly increasing the execution time due to the fixed preprocessing operations, by exploiting the background and facility information. With YOLOv8, YOLOv10, and YOLOv12 small, medium, and large models, however, the proposed method improves both the baseline accuracy and execution time because the portion of the fixed preprocessing operations is relatively decreased with bigger models.Highlights The three graphs present overall performance improvements in accuracy and latency across different YOLO models. The red lines indicate the baseline models, while the blue lines indicate the proposed models. The black lines represent the performance changes for each model, highlighting the improvement from the baseline to the proposed method. What are the main findings? What is the implication of the main finding? With YOLOv8, YOLOv10, and YOLOv12 nano models, the proposed method improves the baseline accuracy, while slightly increasing the execution time due to the fixed preprocessing operations, by exploiting the background and facility information. With YOLOv8, YOLOv10, and YOLOv12 small, medium, and large models, however, the proposed method improves both the baseline accuracy and execution time because the portion of the fixed preprocessing operations is relatively decreased with bigger models.Abstract Standard Deep Learning-based detectors generally face a trade-off between accuracy and latency, as well as a significant performance degradation when applied to unseen environments. To address these challenges, this study proposes a method that enhances both accuracy and latency by leveraging the static characteristics of fixed-camera pig pen monitoring. Specifically, we utilize background and infrastructure information obtained through a one-time preprocessing step upon camera installation. By integrating this information, we introduce three distinct modules, Background-suppressed Image Generator (BIG), Facility Image Generator (FIG), and Background Suppression Integration (BSI), that improve detection accuracy and operational efficiency without the need for model retraining. BIG creates background-suppressed images that integrate foreground and background information. FIG creates facility mask images that can be used to identify pigs that are occluded by facilities, enabling more efficient learning in unseen environments. BSI leverages both the input image and the background-suppressed image generated by BIG, feeding them into a 3D convolution layer for efficient feature fusion. This difference-aware fusion helps the model focus on foreground information and gradually reduce the domain gap. After training on the German pig dataset and testing on the unseen Korean Hadong pig dataset, the proposed method could improve AP50 accuracy (from 75% to 86%) and Jetson Orin Nano latency (from 67 ms to 41 ms) compared to the baseline model YOLOV12m.

키워드