This page was machine-translated and may differ from the original. View original

Forty2Dot: "Our 'Lane Detection Dataset' is Different from Existing Datasets"

Google 우선 소스Published2022.06.16 15:13

▲Song Chang-hyun, CEO of Forty2Dot, is giving a presentation at NextRise 2022. (Photo: Seong Yu-chang)


Available for use by academia, research institutes, startups, developers, etc.
Dataset program '42dot Open Dataset' released

CEO Song Chang-hyun announces participation in NextRise2022 and CES2023.
Following the release of its self-built dataset to contribute to the development of a domestic autonomous driving research and development ecosystem, 42dot announced its participation in CES 2023 at NextRise 2022.

FortiTwoDot announced on the 15th that it will sequentially release datasets for learning artificial intelligence models for autonomous driving through the '42dot Open Dataset' program.

FortyDot's dataset contains images and labeling information acquired from various urban road environments in Korea.

Until now, most domestic autonomous driving research has utilized datasets provided by overseas institutions, but many have pointed out that it is difficult to apply them to domestic roads with many intersections, side roads, and curves.

The company expects that by releasing the high-quality training dataset collected domestically, Forti2Dot will contribute to the development of domestic autonomous driving research capabilities and the cultivation of excellent engineers.

The first thing that Forty2Dot released today was a dataset on lane detection and multi-object tracking.

The lane detection dataset provides approximately 39,000 high-resolution images collected for autonomous driving on domestic highways as well as complex urban roads.

Lanes in complex road environments These are very important road signs for drivers and pedestrians to follow traffic laws and drive safely.

Existing lane recognition studies have been limited to detecting lanes in highway environments or detecting simple straight lanes, which has limited accurate driving guidance.

A Porty2Dot representative said, “The lane recognition dataset will be differentiated from existing datasets as it includes various types of roads, such as curves and intersections.”

FortiTwoDot also releases a multi-camera multi-object tracking dataset (MCMOT).

FortyDot's multi-camera system acquires various road information from three front cameras and provides the type, location, and unique ID of objects appearing in the image.

Autonomous driving systems utilize a combination of sensors by positioning multiple cameras at various angles to perceive the surrounding environment in all directions.

This is similar to how drivers primarily rely on visual information to perceive the road environment.

In particular, to control a vehicle naturally, it is important to simultaneously process image information input to a multi-camera system and integrate and transmit the results.

In the case of a conventional independent camera system, if the object being tracked leaves the field of view of an individual camera, a completely new tracking ID is assigned, losing continuity with previous movements, ultimately generating unnatural vehicle control commands.

On the other hand, multi-camera systems can continuously track the movement of objects, even if they briefly leave the field of view, similar to how a driver can recognize the same object by moving their head or relying on short-term memory.

The dataset being released this time can be used by anyone, including academics, research institutes, startups, and developers, after applying.Capable.

Forty2Dot plans to award winners of the Dataset Challenge and introduce Forty2Dot's autonomous driving technology at a workshop held jointly with the Korean Conference on Computer Vision (KCCV) in August.

Meanwhile, Song Chang-hyun, CEO of Forty2Dot, appeared as a speaker at NextRise2022 on the 16th and presented Forty2Dot's vision and values.

Forty2Dot is providing 'AKit', a full-stack autonomous driving technology that moves on its own, and 'TAP!', a mobility platform that provides travel routes, through its urban integrated solution 'UMOS (Urban Mobility Operating System)'.

"TAP!" has 3,505 registered users and has recorded 652 rides since February 2022. Song said that 20% of those users use it regularly, establishing it as a popular public transportation option.

Additionally, Forty2Dot's autonomous vehicle is not equipped with a lidar. Instead of lidar, we installed seven cameras to ensure maximum safety.

This is the result of the company's efforts to implement technology that makes the exterior of the vehicle look as similar as possible to that of a regular vehicle.

CEO Song Chang-hyun said, “Autonomous driving is no longer a distant dream,” and added, “‘TAP!’ will rapidly expand its business independently from ‘Akit’ in the future.”

He then announced that he would introduce a new autonomous vehicle model in September 2022 and participate in CES 2023, raising expectations that Forty2Dot will play a leading role in domestic autonomous driving technology.
본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.
성유창 기자
성유창 기자