Autonomous Driving
Autolabeling, 4D BEV, and data closed-loop provide full-process data technology support for autonomous driving.
The autonomous driving industry is characterized by large-scale data, diverse scenarios, fast iteration of data needs, lack of unified standards and specifications, and strong demands for cost reduction and efficiency improvement. To quickly adapt to various mass-produced vehicle models and survive and gain an advantage in complex and differentiated competition, the industry's demands for data services focus on higher confidentiality, efficiency, quality, professionalism, and lower costs, interface management costs, and problem feedback time. Since its inception, Stardust Data has been deeply involved in autonomous driving data technology, currently serving more than sixty companies, including autonomous driving algorithms, OEMs, and Tier 1 suppliers. Its capabilities and experience cover the full range of data annotation needs for all types of autonomous driving scenarios. In addition, Stardust also provides data strategy design, vehicle modification, data collection, data management, data preprocessing, model evaluation, data mining, data resource sharing, simulation testing, and real vehicle testing as part of its comprehensive data closed-loop services for autonomous driving.
Application Scenarios
Smart Cockpit
Stardust can collect and annotate data for all types of driver behavior in the cockpit, including but not limited to:
Driver behavior detection data collection and annotation (demonstrated with compliance and de-identification)
Gesture action collection and annotation
Speech command collection, segmentation, and transcription
......
Technical Skills
Cutting-edge annotation scenarios
Support ultra-large point cloud annotation (200M+ points), 4D BEV, and continuous frame sequences (2000+ frames). Enable multi-sensor fusion annotation including LiDAR, radar, and cameras.
Comprehensive annotation tools
Autolabeling Function
Leading Project Management
Corner case detection
Case Studies
Annotation Project: 4D Object Tracking
A Lidar Unicorn Company
Requirement Description: Joint target tracking annotation for 2D and 3D data from roadside equipment. The pain point lies in the lack of accuracy in time alignment and static calibration of the original data, necessitating adjustments to time alignment and camera parameters to ensure accurate mapping.
Solution: Stardust's self-developed 4D annotation tool can annotate targets over a period of time, thereby identifying the time offset between sensors. After aligning the time, Stardust optimizes static calibration parameters through its proprietary calibration algorithm. Translation accuracy can reach the centimeter level for translation and within half a degree for rotation. This addresses the pain point of poor dynamic and static calibration accuracy in the client's original acquisition equipment.
A certain automotive company's service project: Mass production solutions for automatic parking and highway cruising.
Annotation Project:
2D: Lane lines, traffic lights, parking spaces, pedestrian object detection (OD), vehicle object detection (OD), road sign annotation, CarPosition, etc.
3D: Single-frame point clouds, unified point clouds, point cloud semantic segmentation, etc.
Why Stardust
Explore More
Fill out the form to schedule a personalized demo with our team. Experience firsthand how our innovative solutions can meet your needs and drive success.
Copyright © 2024 StardustAI Inc. All rights reserved.