Simplined Technology - Smart City/Mobility -Macnica

Scalable and flexible Ground Truth generation toolchain

What is the "Ground Truth System (GTS)" that supports autonomous driving development?

The Ground Truth System is not simply a collection of sensors.
This system semantically records events occurring in front of the vehicle while driving and outputs them as "correct data" that can be directly used by the AI for learning. The results are output in CSV or JSON format and can be immediately used as training or validation data.

Why is high-precision ground truth required for autonomous driving now?

Are you facing any of the following challenges in your autonomous driving project?

Even with a large amount of data collected, the label quality can vary, preventing the model from training adequately.

Manual labeling is extremely time-consuming and costly, resulting in slow development cycles.

There is a discrepancy between the field data obtained from actual vehicle testing and the data used for training, resulting in false detections in actual operation.

These are not merely operational issues; they directly impact safety. Just as humans can misread lanes or traffic signals, AI judgments become unreliable when the "accuracy" of the data falters. Ground Truth is the foundation for eliminating that uncertainty and creating "real-world correct answers."

Simplined's Ground Truth Generation Technology Realized Through Sensor Fusion and AI

Simplined's Ground Truth System is more than just an automation tool. It's designed with a focus on unifying what happens in the field across the verification, training, and operation phases. Key elements include:

■ Multimodal acquisition (LiDAR + Camera + GNSS/INS)

LiDAR is used to acquire 3D shape and distance (detection is possible at a maximum distance of 500m).
Color information and sign text information are extracted from multiple cameras and overlaid with information obtained from LiDAR.
GNSS/INS ensures consistency of time, azimuth, and position, minimizing blur between point clouds and images.

■ Automatic calibration (no target required)

By automatically optimizing the external parameters of the LiDAR and camera using surrounding structures (terrain and buildings) without the need for calibration targets that were previously required, the burden of on-site setup is significantly reduced.

■ High-precision static and dynamic labeling achieved with AI

Dynamic GT: Generates detection and tracking data (with tracking ID) for vehicles, motorcycles, pedestrians, etc.
Static GT: Semantically records road elements such as lanes, crosswalks, arrows, stop lines, signals, and signs.

■ Multilayer output format

Generates 3D point clouds, color point clouds, BEV maps, and semantic maps.
The final output is exported in CSV/JSON format and can be directly used in AI training pipelines and test evaluations.

The advantages of the Simplined GTS in accelerating autonomous driving development

① Achieving both development quality and practicality through high-precision true value generation and wide-area observation.
GTS achieves a Precision/Recall rate of 98% in major categories (vehicles, pedestrians, etc.), demonstrating high accuracy and reproducibility. Furthermore, it supports wide-area observation up to 500m ahead, providing useful data for high-speed driving environments and predictive algorithms. This reduces the risk of misfitting and overfitting, improving quality in both learning and evaluation.

② Improving development efficiency at the practical level through automation, standardization, and visualization.
Automatic calibration and a universal data standard reduce variations between sites and facilitate data integration across multiple locations. Furthermore, intuitive visualization outputs, such as color information and LiDAR point clouds, allow engineers to quickly identify AI errors and smoothly proceed from root cause analysis to improvement. As a result, labeling time is reduced, and personnel can be concentrated on review and decision-making processes.

A self-developed and continuously expandable white box platform autonomous driving

The data generated by GTS is designed to be used consistently throughout the entire process, from training to testing, verification, and mass production. Furthermore, from the fall of 2026 onwards, we plan to provide a white box (OSS) of Realtime Perception, which is realized through the fusion of LiDAR and camera sensors. This will allow users to modify, expand, and optimize the system in-house according to their own requirements, without relying on a black Box. This will help avoid vendor lock-in and build a development foundation that allows companies to maintain technical leadership in long-term operations.

About Simplined, a company accelerating autonomous driving technology

Simplined focuses on providing comprehensive data solutions for autonomous driving applications. Leveraging data-driven AI technologies, we support the implementation of advanced intelligent driving.
Established in 2023, the company is headquartered in Hong Kong, China. Currently, it operates in Shenzhen (China), Tokyo (Japan), Shanghai (China), Seoul (South Korea), and Stuttgart (Germany).

FAQ

What is the accuracy of the true value? How is it verified?

▼

Farthest: farthest
Nearest: Shortest
The accuracy data above is based on manual annotation and compared to the results calculated by this system.

Can the true readings be trusted even in environments such as rain, nighttime, or under cover?

▼

This system is designed for evaluation in real-world environments and is intended to acquire data under multiple conditions, including clear skies, nighttime, backlighting, and mild adverse weather. However, in cases where observation is difficult due to strong obstruction or sensor physical limitations, the uncertainty of the true values may increase. Therefore, it is intended to be used to "understand the limits of the evaluation target," including under such conditions.

How do you perform time synchronization?

▼

Time synchronization is performed by using the GNSS reference time as a base and referencing protocols such as PPS and PTP to synchronize the time of the LiDAR and camera.

Can this true value data be used directly for algorithm evaluation?

▼

Yes, the system outputs true results in a data format intended for evaluation purposes, and includes 3D position information of the object, tracking ID, time-series information, etc., so that it can be used directly for comparative evaluation with recognition results (Precision/Recall, etc.).

Inquiry

If you have any questions or concerns regarding Simplined Technology, please feel free to contact us.

Site Search