The team behind the system claims it can detect 2D and 3D objects with an accuracy greater than 96%.
An international research team claims it has created a system for self-driving vehicles that can detect objects with high accuracy, making for a safer and more reliable autonomous experience.
The team, led by Professor Gwanggil Jeon from Incheon National University in Korea, has developed a smart Internet-Of-Things (IoT) enabled end-to-end system that can detect objects in real-time, based on deep learning. The system is claimed to be able to detect other 2D and 3D objects with an accuracy greater than 96%.
"For autonomous vehicles, environment perception is critical to [answering] 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," notes Professor Jeon.
"We devised a detection model based on YOLOv3, a well-known identification algorithm. The model was first used for 2D object detection and then modified for 3D objects," he elaborates.
But before we look at the inner workings of the new system, one needs to understand how autonomous vehicle systems work. Self-driving cars rely on light detection, LiDAR, sensors, and RGB cameras that create data as RGB images and 3D measurement points. This needs to be quickly and accurately processed to safely identify pedestrians and other vehicles on the road.
By integrating IoT technology and advanced computing methods into self-driving vehicles, it would allow for quicker data processing and enable autonomous cars to navigate obstacles and environments with greater efficiency. As such, the team fed point cloud data and RGB images into the YOLOv3 identification algorithm. Early results are promising; not only did it outperform other detection models, but it achieved the aforementioned detection accuracy rating of more than 96%.
But is this just some technological trickery reserved for the future, or is there a real-life use for this now?
"At present, autonomous driving is being performed through LiDAR-based image processing, but it is predicted that a general camera will replace the role of LiDAR in the future. As such, the technology used in autonomous vehicles is changing every moment, and we are at the forefront. Based on the development of element technologies, autonomous vehicles with improved safety should be available in the next 5-10 years," said Professor Jeon.
This system can reputedly be applied to autonomous vehicles, parking, delivery, and even robots.
Safety is a big concern for the autonomous future, a point highlighted by Audi's Uta Klawitter earlier this year. Even if it's proven to be convenient and efficient, users want to know it's safe. "Only then - and this is the second challenge - will it gain social acceptance and the corresponding trust," said Klawitter at the time.
In the United States, several companies are trialing autonomous technology, but GM-backed Cruise already operates self-driving taxis in several cities, with plans to expand to more regions in the coming year. It hasn't been a flawless process, though.
Earlier this year, one of the company's self-driving Chevrolet Bolt taxis was involved in a serious collision and, prior to that, a fleet of its vehicles descended upon a busy San Francisco street and refused to move, causing chaos for commuters and residents.
It's clear to see that there's still a lot of work to be done before autonomous vehicles gain the trust of the greater public, and this will be done by developing safety solutions, much like Professor Jeon and his team has. For now, the idea of owning a private autonomous vehicle is some time away - but the technology will slowly become mainstream as more ride-share app companies forge ahead in this field.