#9Recognition sensors required for automated driving③
One of the technologies that advances rapidly today is a camera (an image sensor).The role of cameras is to recognize objects in the same way as humans do, so the key is how to achieve the level of recognition equivalent to humans through programs. In brief, assuming it has high resolution, software that utilizes data obtained by the camera decides the game.
The software includes algorithms that distinguish various objects in the driving environment such as vehicles, lane markers, roads, pedestrians and road signs.To improve recognition performance, we need as many samples as possible.It is critical to improve completeness to cover actual road environments.To achieve the completeness, we need to collect as many and various driving scenes as possible by driving roads all over the world.For example, many of road division lines in Japan are while lines. But it isn’t true for other parts of the world.Some countries use blue lines, some use yellow, and a certain type of raised pavement markers, which is called Bott’s Dotts, are used in North America. All of them have to be recognized in the same way as road division lines.There is a limit for humans to extract characteristics amount from huge amount of collected data, so, deep learning that machines directly extract characteristics is receiving much attention in recent years.
In order to improve accuracy of recognition required for active safety and automated driving, it is effective to use mulitple types of sensors such as laser radar, millimeter-wave radar and vision sensors in combination.That is because, generally speaking, radar can’t recognize the shape of objects, and camera functions are limited under certain conditions, like in the fog. To provide a safe and secure automobile society for all people in the world, we need to improve the performance of respective sensors even more.In addition, combining multiple sensors based on various methods is considered to be an effective method robust to environmental changes.