Real-Time Ground Vehicle Detection in Aerial Infrared Imagery Based on Convolutional Neural Network

• DOI: 10.3390/electronics7060078
• Publication Date: 2018-05-24T06:55:43Z
• Authors: Xiaofei Liu; Tao Yang; Jing Li
• Keywords: Aerial image

An infrared sensor is a commonly used imaging device. Unmanned aerial vehicles, the most promising moving platform, each play vital role in their own field, respectively. However, two devices are seldom combined automatic ground vehicle detection tasks. Therefore, how to make full use of them—especially based on imagery–has aroused wide academic concern. due imagery’s low-resolution and detection’s complexity, extract remarkable features handle pose variations, view changes as well surrounding radiation remains challenge. In fact, these typical abstract extracted by convolutional neural networks more recognizable than engineering features, those complex conditions involved can be learned memorized before. this paper, novel approach towards images network proposed. The UAV application firstly introduced. Then, platform built an dataset unprecedentedly constructed. We publicly release (NPU_CS_UAV_IR_DATA), which for following research field. Next, end-to-end built. With large amounts recognized being iteratively learned, real-time model It has unique ability detect both stationary vehicles real urban environments. evaluate proposed algorithm some low–resolution images. Experiments NPU_CS_UAV_IR_DATA demonstrate that method effective efficient recognize vehicles. Moreover it accomplish task while achieving superior performances leak false alarm ratio.