Rapid Ground Car Detection on Aerial Infrared Images
Xiaofei Liu, Tao Yang, Jing Li, Miao Wang, Yanning Zhang · 2018
With extensive applications of unmanned aircraft vehicle and infrared imagery's particular characteristic, ground car detections using infrared aerial images have been gradually applied to intelligent video surveillance. However, the aerial infrared images are always low-resolution and fuzzy, ground car detection is subjected to pose variations, view changes as well as surrounding radiations, this inevitably poses many challenges to detection task. In this paper, we present a novel approach toward ground car detection on infrared images via an end to end regressive neural network, other than background segmentation or foreground extraction. The main works of our research can be divided into three parts: (1) A unique aerial moving platform is built to collect a large amount of infrared images. It is achieved by assembling the DJI M-100 UAV and the FLTR TAU2 infrared sensor; (2) An aerial infrared car data set is unprecedentedly constructed. It is can be used for the following researches in this field; (3) A ground car detection model is trained. It can work in the moving and stationary cars in some severe environments. We test it on some low-resolution infrared images in a typical urban complicated environment and compare it with a state-of-the-art method. Experimental results demonstrate that the proposed approach instantly detects cars while keeping a low leak and false alarm ratio.