An Infrared Moving Small Object Detection Method Based on Trajectory Growth

Dilong Li, Shuixin Pan, Haopeng Wang, Yueqiang Zhang, Linyu Huang, Hongxi Guo, Xiaolin Liu, Qifeng Yu

Abstract: Aiming at the detection of infrared moving small objects for moving camera, an infrared moving small object detection method based on trajectory growth is proposed. In the first stage, the proposed method firstly estimates the homography between images by pairs of corresponding points obtained by pyramidal Lucas-Kanade optical flow method. Then the changed regions in the image are extracted by the three-frame difference method, and the optimized segmentation method is used to produce accurate extraction results. In the second stage, based on the assumption of short-time approximate uniform linear motion, the old trajectories are grown and the new trajectories are generated to reduce the interference of false alarm. Then, the final detection results are obtained by confirming the trajectories. Experiments on publicly available dataset show that the proposed method achieves F1 score higher than 0.91 on average, indicating it can accurately detect small objects in complex and even noisy background with low false alarm rate and high efficiency.