NEC Corp exhibited a prototype of a system that recognizes 3D objects with a monocular camera at ITS World Congress Tokyo 2013, which took place from Oct 14 to 18, 2013.
Currently, NEC is developing technologies for the system at its research institute. The system can distinguish six kinds of objects and measure their distances and postures.
"We have achieved a distance accuracy equivalent to that of a system using a compound-eye camera," the company said.
The new system can measure distance and posture because the car equipped with the system or another vehicle moves. Specifically, it calculates a distance, etc from differences between video frames. Also, it can distinguish (1) large vehicles including trucks, (2) mid-sized vehicles including minivans, (3) standard-sized vehicles, (4) pedestrians, (5) bicycles and (6) motorbikes.
"We realized the recognition capability by using a machine learning method," NEC said.
Watching a movie of the system recognizing objects, I thought that its recognition accuracy was considerably high.
At this point, the system cannot recognize objects during nighttime hours or in bad weather. And for the calculation, it is necessary to use a multi-core CPU such as those used in personal computers.