In recent years, data labeling with the rapid development of artificial intelligence, intelligent driving , as an important part of strategic emerging industries, has attracted more and more attention. Intelligent driving technology refers to the technology that relies on machines to drive and completely replaces people in special cases. It mainly includes three links: network navigation, autonomous driving and manual intervention. At present, the mainstream algorithm model of autonomous driving is mainly based on the supervised deep learning method, which requires a large amount of structured labeled data to train the model.

It is said that data is the blood of artificial intelligence, and data will only become meaningful if it is marked. Therefore, the vigorous development of artificial intelligence has also promoted the continuous growth of data collection and labeling companies .

 

Types of Intelligent Driving Data Labeling

The common types of data annotation in the field of automatic intelligent driving mainly include vehicle and pedestrian labeling, 3D cube labeling, 3D radar point cloud labeling, sign and signal light labeling, polyline labeling, semantic segmentation, video tracking labeling, etc.

1. Vehicle/pedestrian labeling

Vehicle and pedestrian frame labeling is widely used in the basic recognition of vehicles and pedestrians. Simply put, it is to label vehicles and pedestrians in the picture, and use the attributes of the frame to develop the test model.

 

2D vehicle marking and pedestrian marking realize the basic recognition of intelligent automatic driving technology, that is, marking cyclists, pedestrians, and cars.

Jinglianwen Technology has collected and marked “2D drawing frame labeling training set of 100000 pictures”, “inspection car 2D labeling image training set of 49980 pictures” and other data sets that are directly used for the research of automatic driving algorithms.

 

2. Vehicle polygon labeling

Vehicle polygon annotation can accurately mark the shape information of the vehicle, which can be applied to the identification of vehicle types, such as vans, trucks, buses, cars, etc., to train automatic driving, and to selectively follow cars or change lanes on the road.

Technology has collected and annotated the “23,000 Vehicle Polygon Annotated Image Dataset”, which can be directly used in the research of automatic driving algorithms.

 

3.3D Cube Labeling

3D cube labeling is to further 3D label the vehicles in the 2D pictures, and is mainly used to judge the volume of vehicles passing by during driving.

Technology has collected and annotated the “21,000 Vehicle 3D Cube Annotated Image Dataset”, which can be directly used in the research of automatic driving algorithms.

 

4.3D radar point cloud labeling

3D radar point cloud annotation is to use 3D images to mark the location and size of objects in the video scene. 3D radar point cloud annotation is mainly used in the construction of autonomous driving virtual reality.

3D point cloud continuous frame labeling is a widely used data processing labeling type in autonomous driving scenarios. It has high requirements for three-dimensional space perception and multi-frame simultaneous collaborative processing capabilities. When labeling, it is necessary to pay attention to the continuous frame object ID, object Frame size, labeling standards for soft and rigid objects, etc.

Jinglianwen Technology has collected and annotated the “24,000 3D Point Cloud Annotated Image Training Set of Vehicles and Pedestrians” to annotate the 3D point cloud information data of the road collection pictures, which can be directly provided to algorithm manufacturers for research on automatic driving algorithms.

 

5. Marking signs/signal lights

Labeling of signs and signal lights is a comprehensive labeling of signs and signal lights suspended on the road. The labeling includes area labeling and semantic labeling, so that autonomous driving can drive safely according to traffic rules.

collected and annotated data sets such as “10,000 Signage Annotated Image Dataset”, “15,000 Signal Light Annotated Image Dataset”, etc., which can be directly used in the research of automatic driving algorithms.

6. Lane marking

Lane line labeling is a comprehensive labeling of road ground markings, including classification labeling, area labeling and semantic labeling. It is used to label lane lines in intelligent driving scenarios, so that intelligent driving vehicles can follow the lane lines. driving according to the rules.

 

Technology has collected and annotated the “24,400 Polyline Annotated Image Dataset”, which can be directly used in the research of autonomous driving algorithms.

7. Semantic Segmentation

Semantic segmentation is a relatively broad type of labeling, which is to segment and label different areas in the picture, which can help intelligent driving vehicles identify drivable areas on the road.

Technology has collected and marked the “50,000 Image Semantic Segmentation Dataset”, which can be directly used in the research of autonomous driving algorithms.

 

8. Video object tracking and labeling

Video tracking and labeling refers to tracking and labeling the vehicles driving in the video, and marking the frame according to the frame capture of the picture. The marked photos are then recombined and arranged into video data in sequence, which is used to train the automatic driving model.

Technology has collected and marked the “21,400 Video Tracking and Labeling Dataset”, which can be used for the research of autonomous driving algorithms.

 

9. ASR voice transcription

ASR speech transcription is often used in the field of voice assistants in automatic driving systems, which can well assist drivers in managing and controlling vehicles.

Technology has collected and marked the ” 88,000 Segments of Vehicle Wake-up Word Voice Dataset “, which can be collected in a static environment in the driving car, with the windows closed, the engine on, the air conditioner in 2nd gear, the driver’s position, and the distance between the microphones is 140mm. It is directly used in the algorithm research of human-computer interaction related to autonomous driving, which greatly saves the research and development time of algorithm research manufacturers.

 

One-stop data solution for intelligent driving data labeling

In order to provide underlying data support for the majority of intelligent driving technology R&D manufacturers,  Technology has collected and marked “235,598 DMS Driver Behavior Training Sets”, ” 260,494 License Plate Image Datasets “, “10,000 Semantic Road Semantic Segmentation”, “600 Hourly Hebei Surveillance Video Dataset” and other data sets that can be applied to different technical algorithms in the field of intelligent driving. It can be directly used in algorithm research without further processing, which effectively shortens the time for enterprise research and development.

 

Technical advantages

Jinglianwen Technology has been certified as a high-tech enterprise. It has a professional technical team, a self-developed data labeling platform, a mature labeling, review, and quality inspection mechanism, and supports computer vision (drawing frame labeling, semantic segmentation, 3D point cloud labeling, key Point labeling, line labeling, 2D/3D fusion labeling, target tracking, image classification, etc.), voice engineering (voice cutting, ASR voice transcription, voice emotion judgment, voiceprint recognition labeling, etc.), natural language processing (OCR transcription, Text information extraction, NLU sentence generalization) multi-type data labeling, through pre-labeled intelligent data processing methods can reduce data labeling time and save labor costs.

 

Technology has also accumulated a wealth of labeling experience for projects such as vehicle and pedestrian drawing frame labeling, lane line labeling, and 3D radar point cloud labeling in intelligent driving scenarios. channel and convert the format according to Party A’s requirements, proofread the data, and basically achieve a labeling accuracy rate of 99%, completing high labeling, high quality, and fast delivery. Up to now, Jinglianwen Technology has reached long-term friendly cooperation with many leading car companies in the industry.

Not limited to intelligent driving data solutions, Technology has developed data solutions in many fields where artificial intelligence is implemented, has profound algorithm understanding ability and database design ability, and implements butler service, deploys projects in advance, adopts Professional business and project managers conduct one-on-one connection with customers, can provide services for customers 24 hours a day, and deal with sudden emergencies in a timely manner. Jinglianwen has also formulated a set of data security assurance system, and signed confidentiality agreements with customers and company employees who have been exposed to the project to reduce the risk of data leakage.

 

In the future,  Technology will be committed to becoming a more professional data collection and labeling company, building more high-quality collection and labeling environments, and providing customers with one-stop data solutions. Contact us for data collection and labeling requirements~