Cattle Trembling Detection Using HFR-Video-Based DIC Analysis

This study proposes a high-frame-rate (HFR) video analysis method that functions as a software-based vibration sensor to estimate when, where, and which body parts of cattle exhibit trembling by detecting tens-of-Hertz frequency components. The proposed sensor estimates velocities at multiple points on the cattle with subpixel precision using HFR-video-based digital image correlation, which is combined with a convolutional neural network-based object detection method to update segmented regions in each frame, even when multiple cattle are moving. We validated our proposed method using 1920 × 1080 video captured at 125 fps for multiple juvenile cattle in an indoor barn, demonstrating that the software-based vibration sensor can detect and visualize short-term trembling behavior with frequencies of 10–14 Hz.