#Japan #Tokyo #AIAutomation #Nexdata #AutonomousDriving

Nexdata Enhances Autonomous Driving with New Datasets

Nexdata, one of Japan's largest AI training data providers, has officially launched a set of high-precision learning datasets specifically designed for the development of autonomous driving and Advanced Driver Assistance Systems (ADAS) in Japan. This initiative marks a significant step forward, addressing the pressing need for high-quality and locally relevant training data that has been a persistent challenge in the field of automated driving.

The Challenge of Insufficient Driving Data

In the current landscape of autonomous driving technology, the development of algorithms hinges not just on the quantity but crucially on the quality and region-specific relevance of learning data. Japan presents unique challenges, including its distinct road signage regulations, complex urban environments, and stringent traffic laws. These factors render generic datasets, often collected abroad, insufficient for effectively improving recognition accuracy in real-world conditions.

Furthermore, acquiring multimodal data—comprising 2D images, 3D point clouds, and temporal lane information—requires significant resources that development teams often struggle to allocate, diverting attention from critical algorithm development tasks.

A Real-World Solution: On-Demand Datasets

In response to these challenges, Nexdata has developed a sophisticated multimodal dataset based on real-world data collected from Japan’s actual road environments. Their approach employs a closed-loop system for managing everything from data collection and annotation to quality assurance and delivery, ensuring that developers have immediate access to practical and applicable training data right when they need it.

Dataset Overview

The dataset encompasses data gathered from driving through urban and coastal highways using vehicles equipped with LiDAR point clouds, synchronized RGB cameras from six viewpoints, RTK-GNSS/IMU, and CAN bus signals, all meticulously synchronized at millisecond intervals. It contains comprehensive annotations for high-precision 3D object tracking, 4D lane recognition, and 2D traffic sign detection, catering to enhancing the learning efficiency of environmental recognition models and the precision testing of object tracking algorithms, as well as supporting HD map construction and ADAS functionality validation.

Annotation Details

1. 2D Traffic Signs
The dataset incorporates a wide array of regulation, instruction, and warning signs used across Japan, featuring bilingual labels in both Japanese and English (e.g., '通行止め' / 'road_closed'). It comes with high-precision 2D bounding boxes along with class attributes and status flags (such as night visibility, dirt, and obstructions). Images are provided in 4K resolution (4096×2160), with annotations delivered in JSON format.

2. 3D Bounding Boxes
Using six 4K onboard cameras and 16-line LiDAR with time-space synchronization recorded at microsecond levels, this portion of the dataset is equipped with three-dimensional bounding box annotations for vehicles, pedestrians, two-wheelers, and obstacles. It includes external and internal parameter calibration files (calib.yaml), enabling immediate initiation of sensor fusion model training.

3. 4D Lane Recognition
Unlike conventional single-frame data, this dataset provides a time-series point cloud map achieved by accurately overlapping multiple LiDAR frames. It vividly reproduces the shapes of lane markings in a continuous manner, leveraging intensity information to enhance learning efficiency for lane detection algorithms. Support for vector data generation for high-definition maps is also included.

Use Cases

The practical applications for Nexdata's dataset span numerous areas, including:

• - Development of autonomous driving algorithms (object detection, lane recognition, sign comprehension)
• - Validation of ADAS functionalities (specific test data for assessing AEB, LKA, ACC, etc.)
• - Construction of simulation environments (recreating virtual spaces based on real-world data)

Customization Options

Moreover, the datasets can be tailored according to clients' specific development needs and testing scenarios. Customizations can include recording areas, weather conditions, annotation items, and data formats, ensuring that Nexdata can offer optimized training data solutions, even for unique autonomous driving systems and next-generation mobility services.

About Datatang Inc.

• - Company Name: Datatang Inc.
• - Brand Name: Nexdata
• - Location: Chiyoda-ku, Kanda Awajicho 2-105, WATERRAS Annex 6F, Tokyo, Japan
• - Established: February 2020
• - Capital: 500 million yen
• - Business Overview: AI learning data provision services (in-house data and customized data), data collection and annotation services, platform provision
• - Website: Nexdata