TY - CONF

T1 - Tracking Small Objects in Global Motion Conditions for an Ornithological Monitoring System

AU - Obukhova, N.

AU - Motyko, A.

AU - Pozdeev, A.

AU - Savelev, A.

AU - Baranov, P.

AU - Smirnov, K.

AU - Sharivzyanov, D.

AU - Samarin, A.

AU - Kotenko, E.

N1 - Export Date: 23 February 2026; Cited By: 0; Correspondence Address: N. Obukhova; St. Petersburg Electrotechnical University "LETI", St. Petersburg, Russian Federation; email: naobukhova@etu.ru; Conference name: 38th Conference of Open Innovations Association, FRUCT 2025; Conference location: Hybrid, Helsinki; Conference date: 2025-11-05 through 2025-11-07

PY - 2025

Y1 - 2025

N2 - In ornithological video monitoring tasks aimed at studying bird behavior, estimating population size, and tracking migration routes, a major challenge is the robust tracking of targets under global camera motion. Such motion, often caused by pan-tilt or mobile platforms, introduces significant distortions in the optical flow. At the same time, the tracked objects are typically small, low-contrast, and highly dynamic, which considerably reduces the robustness of conventional tracking methods. This study aims to develop and experimentally validate a tracking method that can operate in video sequences affected by global motion, while maintaining high accuracy and real-time performance. The proposed approach integrates a neural network-based tracker and trajectory prediction using a Kalman filter. The method was evaluated on a dataset simulating real ornithological monitoring scenarios, including highly detailed and dynamic backgrounds, moving cameras, variable lighting, and complex object trajectories. Experimental results showed that the tracking failure rate did not exceed 5×10-4, while the average processing speed reached 21 frames per second. Compared to a conventional tracking method based on HOG+KCF and Kalman filtering, the proposed method achieved a 4-fold reduction in tracking failure rate and a 2.5-fold reduction in tracking failures under occlusion conditions. The developed method is designed for use in bird monitoring systems operating in natural and agricultural landscapes, where reliable object tracking is required in visually complex environments. The results demonstrate the potential of the proposed solution for both scientific and ornithological research, as well as applied environmental monitoring tasks. © 2025 FRUCT Oy.

AB - In ornithological video monitoring tasks aimed at studying bird behavior, estimating population size, and tracking migration routes, a major challenge is the robust tracking of targets under global camera motion. Such motion, often caused by pan-tilt or mobile platforms, introduces significant distortions in the optical flow. At the same time, the tracked objects are typically small, low-contrast, and highly dynamic, which considerably reduces the robustness of conventional tracking methods. This study aims to develop and experimentally validate a tracking method that can operate in video sequences affected by global motion, while maintaining high accuracy and real-time performance. The proposed approach integrates a neural network-based tracker and trajectory prediction using a Kalman filter. The method was evaluated on a dataset simulating real ornithological monitoring scenarios, including highly detailed and dynamic backgrounds, moving cameras, variable lighting, and complex object trajectories. Experimental results showed that the tracking failure rate did not exceed 5×10-4, while the average processing speed reached 21 frames per second. Compared to a conventional tracking method based on HOG+KCF and Kalman filtering, the proposed method achieved a 4-fold reduction in tracking failure rate and a 2.5-fold reduction in tracking failures under occlusion conditions. The developed method is designed for use in bird monitoring systems operating in natural and agricultural landscapes, where reliable object tracking is required in visually complex environments. The results demonstrate the potential of the proposed solution for both scientific and ornithological research, as well as applied environmental monitoring tasks. © 2025 FRUCT Oy.

KW - Birds

KW - Cameras

KW - Complex networks

KW - Environmental monitoring

KW - Kalman filters

KW - Motion tracking

KW - Object tracking

KW - Population statistics

KW - Time and motion study

KW - Video recording

KW - % reductions

KW - Failure rate

KW - Global motion

KW - Monitoring system

KW - Monitoring tasks

KW - Motion conditions

KW - Small objects

KW - Tracking failure

KW - Tracking method

KW - Video monitoring

KW - Failure analysis

U2 - 10.23919/FRUCT67853.2025.11239289

DO - 10.23919/FRUCT67853.2025.11239289

M3 - материалы

SP - 248

EP - 254

Y2 - 5 November 2025 through 7 November 2025

ER -