Automatic driving

Self-driving cars, which can also be referred to as driverless cars, are a new type of technology that is realized through control by a computer. Autonomous driving is a combination of artificial intelligence, computer vision, radar systems, global positioning systems and other technologies that can realize safe driving of vehicles without human control.

Automatic driving technology is a very complex and difficult project, because the driving of a motor vehicle is itself a task of high precision, requiring the driver to stay awake at all times and clearly observe the surrounding situation of the vehicle. Due to the complexity of the factors involved in traffic, the driver not only needs to observe the traffic lights and the surrounding vehicles, but also needs to take into account the pedestrians on the road, bicycles, electric vehicles, whether there are obstacles on the road ahead, and even animals that suddenly break into the traffic flow, the road information changes in real time, and the slightest carelessness can lead to traffic accidents.

Self-driving cars received widespread attention as early as 2012, when Google’s self-driving car received the first U.S. license for self-driving vehicles in May of that year. Due to the obvious advantage of sparsely populated areas abroad, self-driving technology has developed more smoothly than at home. Baidu and BMW’s self-driving research program officially opened in 2014 and quickly rolled out prototypes.

Automatic driving involves a variety of technologies, of which a very important part is computer vision, because the vehicle driving needs to use all the time with the eyes to observe all the elements involved in the traffic, so computer vision plays a great role in it, the computer’s high arithmetic power and low labor costs for automatic driving provides a solid foundation.

Traditional target detection feature extraction methods have achieved good recognition results for different targets in traffic scenes, including vehicles, pedestrians, and road surfaces. Automatic driving involves multiple aspects of object recognition, the most basic of which is the recognition of vehicles and roads. Traditional feature extraction methods process and calculate the gray value of lane lines, road edge boundaries, and texture features to segment various regions of the road, but the limitations are large. Since the video and images of the road are often affected by light, obstacles, shadows of trees, cluttered vehicles and pedestrians on the roadside, the traditional simple feature detection methods are difficult to realize the recognition task in complex road conditions. Computer vision technology for autonomous driving has undergone a long iteration of updates from traditional feature extraction methods to computer vision methods using deep learning.

The rise of deep learning has led to a great leap in the quality of accomplishing target recognition detection tasks, in many cases even surpassing humans in terms of accuracy and speed. Deep learning target detection is not just based on the localization of features on the surface of the target in an image compared to traditional detection, but rather in-depth autonomous learning.

Deep learning-based autonomous driving can greatly improve reaction speed by directly learning the correct driving process to perceive the driving method of the actual driving road, and making an overall judgment of the road conditions and targets on the driving road, rather than localized calculations of the road surface, vehicles, pedestrians, and so on separately. Computer vision tasks in autonomous driving also include many kinds, such as: vehicle localization, 3D visual reconstruction, object detection and classification, semantic segmentation, instance segmentation, panoramic segmentation, motion estimation, situational reasoning, uncertainty reasoning and so on. Therefore, autonomous driving is a complex and complex project, and more thorough and rigorous testing is needed to ensure absolutely safe driving.

Your task

What other scenarios do you know that utilize automated driving? This can be illustrated with an example.

Share your thoughts and ideas in the comments below.