Autonomous Driving Data
Elevate your autonomous driving projects with SurfingTech’s specialized data services. Offering advanced point cloud annotation and comprehensive 2D & 3D data annotation solutions,we cater to the intricate needs of autonomous vehicle training. Our precision-driven approach ensures high-quality data for accurate model training, enhancing the safety and efficiency of self-driving technologies. Partner with us for reliable autonomous driving data solutions that drive innovation and progress in the field of autonomous vehicles.
Our Global Street View Collection project aims to meet this demand by providing a state-of-the-art solution for capturing, processing, and delivering high-quality street-level imagery across all corners of the world.
Utilizing high-resolution cameras and innovative capture techniques to ensure every detail is vividly documented.
Covering urban, suburban, and rural areas across all five continents to provide an extensive and inclusive dataset.
Ensuring compatibility with major mapping services, geographic information systems (GIS), and virtual reality platforms.
Offering an intuitive and user-friendly interface for easy access and navigation of street view data.
2D annotation includes bounding box, polygon, semantic segmentation and target tracing.
The 3D annotation plays an important role in improving the accuracy of target and direction detection. Meanwhile, the target detection algorithm can predict the position and attitude of vehicles in real 3D space.
The 3D point cloud annotation platform is developed and owned by SurfingTech. It has two versions: Discrete-image Annotation and Video-sequence Annotation.
The discrete image refers to single frame point cloud data with no time sequential connections. Each frame image needs to be annotated separately as shown below. Very often, the number of points in a single frame image is insufficient, thus it is necessary to project the color of the 2D image onto the corresponding point in the 3D point cloud data. At the same time, annotators need to refer to the 2D image to finish the annotation.
For Video-sequence Annotation (scene-based), we reconstruct the 3D scene using to the continuity between frames first and then annotate it in the 3D scene. The annotation is projected back to each frame of 2D images and point cloud data to form the annotation result.
– Calculate the Transformation Parameter so that points in one sensor can be transformed into the coordinate system of another sensor – Calculate the correspondence between points in 3D point cloud data and points in 2D image data, which is more helpful for improving the accuracy of scene understanding.
Detecting the position area of the calibration plate in Range Data.
Introduce more information to optimize the results of Global Registration.
The final value of the transformed parameter is calculated using a non-maximum suppression method.
Calculate a set of possible Transformation Parameter parameters at a faster rate.
We provide a line-controlled retrofit solution for Lincoln MKZ Hybrid and Geely Borui GE models. It can provide vertical control (throttle, brake) and lateral control (steering system) capabilities, as well as control of gears and turn signals. At the same time, it can provide necessary body state information feedback, such as vehicle speed, wheel speed, steering wheel angle, throttle opening, GPS, four-door two-cover switch status. The modification cycle takes one day, and the company’s wire-controlled modification is a non-destructive modification that does not change the original vehicle’s actuator.