🤖 AI Summary
Existing pose estimation and tracking methods (e.g., STEP, ViTPose) suffer significant performance degradation on caged-animal images due to systematic occlusion caused by cage-bar structures.
Method: We propose a three-stage preprocessing framework: (1) a Gabor-enhanced ResNet-UNet network incorporating 72-directional Gabor kernels for high-precision, multi-orientation cage-bar segmentation; (2) a content-aware inpainting model (CRFill) to remove cage regions; and (3) standard pose estimation and tracking applied on the inpainted images.
Contribution/Results: This framework effectively mitigates occlusion-induced artifacts, enabling keypoint detection accuracy approaching that of unoccluded conditions and substantially improving trajectory consistency. Experiments demonstrate that our preprocessing paradigm significantly enhances the robustness and reliability of animal behavior analysis under complex, structured occlusions.
📝 Abstract
Animal tracking and pose estimation systems, such as STEP (Simultaneous Tracking and Pose Estimation) and ViTPose, experience substantial performance drops when processing images and videos with cage structures and systematic occlusions. We present a three-stage preprocessing pipeline that addresses this limitation through: (1) cage segmentation using a Gabor-enhanced ResNet-UNet architecture with tunable orientation filters, (2) cage inpainting using CRFill for content-aware reconstruction of occluded regions, and (3) evaluation of pose estimation and tracking on the uncaged frames. Our Gabor-enhanced segmentation model leverages orientation-aware features with 72 directional kernels to accurately identify and segment cage structures that severely impair the performance of existing methods. Experimental validation demonstrates that removing cage occlusions through our pipeline enables pose estimation and tracking performance comparable to that in environments without occlusions. We also observe significant improvements in keypoint detection accuracy and trajectory consistency.