Article Preview
Top1. Introduction
With the technological advancements in 5G/6G basic networks (Kumar et al., 2021), fog computing (Al-Qerem et al., 2020), cloud computing (Peñalvo et al., 2022; Vijayakumar et al., 2022), big data (Stergiou et al., 2021), social network (Almomani et al., 2022; Zhang et al., 2023; Arowolo et al., 2023), information security (Gaurav et al., 2023; Alhalabi et al., 2023), Internet of Things (IoT) (Memos et al., 2018), Internet of Vehicles (IoV) (Prathiba et al., 2021; Sharma et al., 2022), smart electric vehicles (Akl et al. 2021), the implementation and utilization of semantic Web-based video surveillance systems have become widespread, resulting in massive video and image data. Vehicle detection technology in semantic Web-based video surveillance systems can provide strong decision support for managers. Compared to other detection technologies (Zhao et al., 2022; Ding et al., 2022; Zhang et al., 2021), semantic Web-based video surveillance has several advantages, including convenient installation and maintenance, no need to interrupt traffic, low cost, large amounts of analyzable information, and no impact on road life. Currently, this technology is experiencing rapid development (Tsai & Chen, 2021; Mo et al., 2022; Zhang et al., 2023).
Vehicle detection is mainly divided into daytime and nighttime vehicle detection (Cui et al., 2022). Due to poor lighting conditions and more complex scenes at night, less information about vehicles and the environment can be obtained. Therefore, the difficulty of nighttime vehicle detection is greater than that of daytime detection (Shao et al., 2021; Zhang et al., 2022; Bell et al., 2022). Numerous vehicle detection algorithms applied in normal weather conditions have limited performance at night. Therefore, there is an urgent need for a high-performance night vehicle detection algorithm (Yang, 2020; Alam et al., 2022; Li et al., 2022).
The traditional vehicle detection algorithms are based on a single light feature for detection, which is difficult to accurately detect with limited feature information (Dai, 2019; Zaarane et al., 2019). The existing deep learning-based detection algorithms are based on multiple features for detection. Hence, these algorithms are significantly superior to traditional detection algorithms in terms of detection accuracy and real-time performance (Alcantarilla et al., 2011; Parvin et al., 2021). However, due to the significant impact of lighting, the potential for misjudgment remains high, and achieving the necessary levels of detection precision and real-time performance to meet the requirements for fast and accurate nighttime vehicles continues to pose challenges (Shao et al., 2021; Huang et al., 2021).