A Novel Background Estimation Method for Infrared Maritime Object Detection
This paper presents a novel approach for fast detection of small objects in complex infrared (IR) images. The main idea is background estimation based on the local minimum filter (LMF). It is inspired by the observation that the intensity of background pixels has similar variations in the local regions and successive frames. Then, a fast localization strategy is proposed for detecting objects from the saliency maps, which are obtained by background subtraction. To evaluate our algorithm, we firstly introduce an open dataset with more than 3, 700 images manually labeled, and then we propose improved evaluation metrics which are more suitable for objects of various scales. Comparing with, most background estimation based methods, our algorithm .shows a better performance with higher precision on large recall values.
infrared objects background estimation dataset evaluation metric
Chongyang Wei Tao Wu Baojun Qi Bo Zhang Qixu Liu
College of Mechatronics Engineering and Automation, National University of Defense Technology, Chang Trucks and Ships Military Delegate Bureau,General Armament Ministry Changsha, Hunan, P. R. China, 41
国际会议
重庆
英文
849-852
2011-08-20（万方平台首次上网日期，不代表论文的发表时间）