TIRNet: Object detection in thermal infrared images for autonomous driving
- PDF / 4,247,382 Bytes
- 18 Pages / 595.224 x 790.955 pts Page_size
- 4 Downloads / 291 Views
TIRNet: Object detection in thermal infrared images for autonomous driving Xuerui Dai1
· Xue Yuan1 · Xueye Wei1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract In the present study, towards reliable and efficient object detection in thermal infrared (TIR) images, we put forward a novel object detection approach, termed TIRNet, which is built upon convolutional neural network (CNN). Instead of using the deep CNN backbone (ResNet, ResNeXt) which suffers low speed and high computational cost, the lightweight feature extractor (VGG) is adopted. To get the robust and discriminating features for accurate box regression and classification, the Residual Branch is introduced. More uniquely, it only exists in the training phase, so no any additional time is increased when inference. All the computation is encapsulated in a single network, so our TIRNet can be optimized and tested in the manner of end-to-end. Furthermore, the continuous information fusion strategy is proposed for improving detection performance, which can effectively solve the problems such as complex background, occlusion, and get more accurate and smoother detection results. To get the real-world dataset and effectively evaluate the effectiveness, a China Thermal Infrared (CTIR) dataset is collected. Besides, we also evaluate our proposed approach on the public KAIST Multispectral dataset. As demonstrated in the comparative experiments, our approach gets the state-of-the-art detection accuracy while maintains high detection efficiency. Keywords Thermal infrared images · Object detection · CNN
1 Introduction Object detection is an irreplaceable role in advanced driver assistance system (ADAS) and autonomous driving [1]. Real-time object detection helps to protect property and life safety [2]. Great progress has been made in the last decades [3, 4]. However, most of the research endeavors are restricted in color images [5]. In fact, some excellent visible cues can be provided by thermal infrared (TIR) imaging sensors [6] rather than color imaging sensors [7]. In the past, TIR images are mainly used in medical, security and military applications. Recently, with lower price and higher image quality, TIR sensors are used for office and home monitoring system, and automotive environment perception [8–10]. Usually, there are two types of infrared sensors are used for automobile applications [6], one is near infrared Xue Yuan
[email protected] 1
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, China
camera (the infrared band: 0.75∼ 1.3μm), and the other is long-wavelength infrared camera (7.5∼ 13μm) which also called TIR camera. The radiation of human body is about 9.3μm [11], so human can be captured robustly by TIR cameras. What’s more, even under the complex environment, TIR cameras are robust and useful. As shown in Fig. 1, TIR cameras are rarely influenced by surrounding lighting changes, the images of high quality can be acquired in the dark, fog and other complex environments. Howe
Data Loading...
