Article Preview
TopIntroduction
In the last years, AI has become an important area of research in many fields: Engineering, science, education, medicine, business, accounting, finance, marketing, economics, stock market, and law, among others (Vrontis, D et al., 2022). The range of AI has grown enormously since the intelligence of machines, especially with machine learning capabilities, has created profound impacts on business, governments, and society. Indeed, they influence the larger trends in global sustainability. Artificial intelligence can be useful to solve critical issues for sustainable manufacturing (e.g., optimization of energy resources, logistics, supply chain management, waste management, etc.). In this context, event detection has been a high domain for AI. Machine learning is one of the most exciting recent technologies in Artificial Intelligence. It is an evolving department of computational algorithms which are designed to emulate human intelligence by learning from the encircling environment. In the past decade, WSN has increasingly adopted advanced machine learning techniques. With advances of deep learning, video surveillance has received a lot of attention and is a major research topic in computer vision (Xu, J,2021). In particular convolution neural networks (CNN), in computer vision applications, the accuracy of classification and object recognition has reached an impressive improvement. The expansion of Graphic Processing Units (GPUs) also outstandingly contributed to the utilization of CNN in computer vision to overcome the problems of real-time processing of computation in complex tasks. In addition, latest trends in cloud robotics have also enabled offloading heavy computations, such as video stream analysis, to the cloud. This allows the processing of video streams in real-time using advanced deep learning algorithms in the context of surveillance applications. Object detection is a general term used to describe a collection of related computer vision tasks that involve recognizing objects in digital photos. Image classification involves predicting the category of an object in an image. Object localization refers to identifying the position of one or more objects in an image and drawing a large number of boxes within its range.
Object detection combines these two tasks, and locates and classifies one or more objects in the image. For many reasons, we choose to use deep learning methods for vehicle detection tasks. First, because deep learning methods can process data in raw format, they are easier to deploy and have better scalability than traditional machine learning methods (LeCun, Y et al.,2015). Moreover, deep learning methods can learn more features that are complex by using multi-level representations (LeCun, Y et al.,2015). Therefore, humans learn these characteristics systematically, not manually.
Furthermore, it is important to show that automatic vehicle detection is indeed useful for modern surveillance camera systems to avoid road accidents (Vadhwani, D et al.,2021) and to organize traffic. So, traffic prediction is the main goal of concern. There are two types of algorithm of object detection: Object detection algorithms using regression includes SSD (Liu, W et al.,2016), YOLO and Object detection algorithms using region proposal includes Faster R-CNN (Ren, S et al.,2015). Hence, this paper aims to evaluate the performance of these algorithms in the context of vehicle detection. It also proposed an architecture of Multimedia Wireless Sensor Network (MWSN) which has the role of capturing vehicles within a tunnel located in the city. With regard to the rest of the paper, Section 2 discusses related works, which treat algorithms used for vehicle detection. Section 3 provides an overview of the Faster R-CNN model, the YOLO model and SSD model. Section 4 presents an optimal architecture of MWSN that can be applied to monitor vehicles. Section 5 presents the performance evaluation of the algorithms for vehicle detection. Section 6 concludes the paper and discusses the main results.