Automated Behavior Recognition and Tracking of Group-Housed Pigs with an Improved DeepSORT Method

Abstract

Pig behavior recognition and tracking in group-housed livestock are effective aids for health and welfare monitoring in commercial settings. However, due to demanding farm conditions, the targets in the pig videos are heavily occluded and overlapped, and there are illumination changes, which cause error switches of pig identify (ID) in the tracking process and decrease the tracking quality. To solve these problems, this study proposed an improved DeepSORT algorithm for object tracking, which contained three processes. Firstly, two detectors, YOLOX-S and YOLO v5s, were developed to detect pig targets and classify four types of pig behaviors including lying, eating, standing, and other. Then, the improved DeepSORT was developed for pig behavior tracking and reducing error changes of pig ID by improving trajectory processing and data association. Finally, we established the public dataset annotation of group-housed pigs, with 3600 images in a total from 12 videos, which were suitable for pig tracking applications. The advantage of our method includes two aspects. One is that the trajectory processing and data association are improved by aiming at pig-specific scenarios, which are indoor scenes, and the number of pig target objects is stable. This improvement reduces the error switches of pig ID and enhances the stability of the tracking. The other is that the behavior classification information from the detectors is introduced into the tracking algorithm for behavior tracking. In the experiments of pig detection and behavior recognition, the YOLO v5s and YOLOX-S detectors achieved a high precision rate of 99.4% and 98.43%, a recall rate of 99% and 99.23, and a mean average precision (mAP) rate of 99.50% and 99.23%, respectively, with an AP.5:.95 of 89.3% and 87%. In the experiments of pig behavior tracking, the improved DeepSORT algorithm based on YOLOX-S obtained multi-object tracking accuracy (MOTA), ID switches (IDs), and IDF1 of 98.6%,15, and 95.7%, respectively. Compared with DeepSORT, it improved by 1.8% and 6.8% in MOTA and IDF1, respectively, and IDs had a significant decrease, with a decline of 80%. These experiments demonstrate that the improved DeepSORT can achieve pig behavior tracking with stable ID values under commercial conditions and provide scalable technical support for contactless automated pig monitoring.

Pig Health Abnormality Detection Based on Behavior Patterns in Activity Periods using Deep Learning

“…This will cause the identification switch errors and increase the ID beyond the real pig's numbers, making the behavior extraction task for individual pigs not stable. To solve this problem, we employed the improved DeepSORT algorithm proposed by S. Tu et al [16], in which we added an additional re-matching step for lost and new tracks using both trajectory and data association processing. This remarkably improves the tracking performance and reduce the identification errors as described in the tracking results section.…”

“…One of the most challenging issues in pig tracking is identification errors caused by identity switching or changing problems during the multi-object tracking process. Some previous works have tried to improve this problem by using additional methods, such as a correlation filter-based tracker via a novel hierarchical data association algorithm [15] or trajectory processing and data association [16]. However, due to the natural conditions of commercial pig farms, such as high www.ijacsa.thesai.org pig density, low light, similar appearances of pigs, or wide covering area of cameras, tracking each individual pig for a long time with low identification error rate is still a difficult task.…”

High-precision tracking and positioning for monitoring Holstein cattle

“…Additionally, it can explore the relationships among distinct targets simultaneously, ultimately generating continuous motion trajectories for each target. Given its mission-specific characteristics, MOT technology holds remarkable practical significance in distinct fields like autonomous driving [ 13 ] and animal monitoring [ 14 ]. Currently, deep learning-based MOT technology can be broadly categorized into Tracking by Detection (TBD) and Joint Detection Tracking (JDT), depending on whether the algorithm framework is end-to-end [ 15 ].…”

Don’t get lost in the crowd: Graph convolutional network for online animal tracking in dense groups

“…DeepSORT (Wojke et al, 2017) is a modern variation of SORT that leverages deep learning by using a wide residual network to extract features related to an objects appearance, which can improve re-identification. SORT and DeepSORT have been applied to group-housed pigs in (Cowton et al, 2019;Shirke et al, 2021;Tu et al, 2022;van der Zande et al, 2021) and (Cowton et al, 2019;Tu et al, 2022) respectively. While DeepSORT is the current state-of-the-art for animal tracking, the performance for long term animal tracking is limited (Tu et al, 2022) and it can perform well only in scenarios in which the detector has few false positives and false negatives (Tu et al, 2022).…”