For self-driving cars to become widely adopted and accepted, designers need to implement efficient, effective, and accurate detection systems. One means to achieve this is through an end-to-end neural network designed to operate with Internet-of-Things technology.
This comes in the form of a recent investigation into the means to improve the safety of self-driving cars using a deep learning-based object detection system. Object detection is based on different algorithms that collapse to perform the recognition and localization of objects. The algorithms utilise deep learning to generate meaningful results. This approach has been undertaken by researchers based at Incheon National University.
In trials, the system can detect object with high accuracy (shown to be above 96 percent) in both 2D and 3D. In comparison with current 2D and 3D detection systems for autonomous vehicles, the new approach is said to deliver better performance.
Self-driving vehicles are equipped with high-technology sensors, including Light Detection and Ranging (LiDaR), radar, and RGB cameras. These monitors collect large amounts of data referred to as measurement points (the “point cloud.”) The quick and accurate processing and interpretation of this collected information is necessary for safety. For example, with the identification of pedestrians and other vehicles.
The information processing occurs through the integration of the Internet-of-Things (IoT) into these vehicles.
To enable all of this to come together effectively, the researchers constructed a detection model based on YOLOv3, which is ab identification algorithm. YOLO represents the phrase “You Only Look Once” and the ‘v3’ reference the third version of the algorithm. YOLO is a Convolutional Neural Network (CNN) designed for performing object detection in real-time. In general, CNNs are classifier-based systems that can process input images as structured arrays of data and recognize patterns between them.
According to lead researcher Gwanggil Jeon: “For autonomous vehicles, environment perception is critical to answer a core question, ‘What is around me?’ It is essential that an autonomous vehicle can effectively and accurately understand its surrounding conditions and environments in order to perform a responsive action.”
Jeon expects autonomous vehicles to be commonplace on roads by 2032.
The technology is presented in the journal IEEE Transactions on Intelligent Transportation Systems, titled “A Smart IoT Enabled End-to-End 3D Object Detection System for Autonomous Vehicles.”
