Recently, AI technology’s application covers many fields, from smart security to autonomous driving. And behind every achievement is inseparable from strong data support. As the core factor of AI algorithm, datasets aren’t just the basis for model training, but also the key factor for improving mode performance, By continuously collecting and labeling various datasets, developer can accomplish application with more smarter, efficient system.

Face Recognition System

For the first time in the Olympics, an facial recognition system was used, which is mainly applied to identify the participants of the Olympic Games, including athletes, volunteers, media, and staff. The system can be achieved that even if many people pass quickly, the photo comparison and authentication can be completed smoothly and automatically. Due to objective factors such as the epidemic situation and summer temperature, the use of face recognition technology can speed up the process of identity verification and entry.

Face recognition technology is based on human facial features. For the input face image or video stream, first determine whether it is a face. If it is a face, then further give the position, size and location information of facial organs. Then generate a code unique to the individual and based on this information, further extract the identity features contained in each face, and compare it with the face features in the data base to complete the authentification. The facial recognition system requires large amount of labeled face data to train the models, and continuously optimizes the models, to achieve a reliable face recognition system with smaller errors.

Face Recognition Data Solution

Nexdata has developed face recognition datasets “23,110 People Multi-race and Multi-pose Face Images Data”, “92,406 People Multi-race and Multi-pose Face Images Data” and “25,949 People Face Recognition Data with Identification Photos”, to help to optimize the face recognition AI models.

Total 23,110 people, includes black people, Caucasian people, Asian, brown people and Indian people. At least 29 images per person. The labeling accuracy rate is 97%.
Intelligent Driving Car

The application of intelligent driving cars during the Olympic Games is also a great innovation. For example, the L4 level e-Palette self-driving minibus is used as a circulating bus in the Olympic Village to transport the personnel of the Olympic Games. Toyota Concept-i, which also has L4 intelligent driving capability, is the parade of the Olympic flame and the pilot car for the marathon game.

Intelligent driving cars perceives the road environment through the on-board sensor system, automatically plan driving routes and control the vehicle to reach a predetermined goal. It uses on-board sensors to perceive the surrounding environment of the vehicle, and controls the steering and speed of the vehicle based on the road, vehicle location, and obstacle information obtained from the perception system, so that the vehicle can drive safely and reliably on the road.

Intelligent Driving Data Solution

In the field of voice interaction and visual perception, Nexdata can provide copyrighted datasets, customized data collection and labeling services, and privatized deployable data labeling platform. We hope to help customers improve the AI models of intelligent driving through high-quality training data to create a safer and more comfortable driving experience.

Street Scene Semantic Segmentation Dataset

The dataset includes more than 10,000 photos and more than 300,000 frames of street view object semantic segmentation data. The data was collected in urban and rural road scenes in China. In terms of labeling, five types of objects (28 sub-categories) on street view images are labeled with polygonal boxes.
AI Scoring Support System

This Olympic Games adopted an artificial intelligence scoring support system. The system performs motion tracking by projecting infrared rays on the player’s body and its surroundings. It will make object detection, object tracking and object recognition based on the motion sequence prediction. In this way, each athlete can be provided with professional scoring result and help them in adjusting their sports posture.

In the process of machine learning, we input the previous competition data to the machine to learn the various movements of the athletes, and the machine will compare it with the learned data in the official competition to score.

Tailored Data Service

Nexdata can provide professional customized data services to meet the customers’ different requirements. Nexdata is equipped with professional data collection devices, tools and environments, as well as experienced project managers in data collection and quality control, so that we can meet the data collection requirements in various scenarios and types. In terms of data labeling, Nexdata has 3 mega-data bases and more than 5,000 professional annotators, supporting on-demand data annotation services, such as speech, image, video, point cloud and text, etc.

Speech Translation Terminal

Panasonic has developed a small terminal device with speech translation function for this Olympic Games, which can translate more than 10 languages. In the demonstration of the translator, there was such an interesting scene: After a foreign man lost his wallet in the stadium, the camera equipped with facial recognition function spotted the man, and the volunteers tried to talk with the man by the multilingual speech translator and finally succeeded in helping him find his wallet.

The process of real-time speech translation is relatively complicated and requires five steps: speech recognition, language understanding, dialogue management, language generation, and speech synthesis. The process of artificial intelligence training is also relatively complicated. First, after entering various prepared speech recognition data into the artificial intelligence learning system, the system will build a statistical model in the words involved in these dialogues and environments. When the user speaks, the software will look for similar content in the statistical model, and then apply it to the pre-”learned” conversion program, so that the audio is converted to text and then from text to speech again, so as to realize real-time translation.

Speech Recognition Data Solution

Nexdata has off-the-shelf 60,000 hours speech recognition data in 17 languages, such as English, French, Spanish, Russian etc. We hope that our high-quality speech data could help our customers in the research of speech recognition.

201 Hours — North American English Speech Data by Mobile Phone and PC

The dataset is collected from more than 300 native speakers in North America. The scene is rich and close to life. The recording environment is quiet room. The recognition accuracy rate is 95%.

Tokyo Olympics has just come to an end. With the development of science and technology, artificial intelligence has begun to shine in all aspects of our life, such as security, medical care, and social services. With the drive of data, we look forward to seeing more AI technologies in the next Olympic Games.

End

If you need data services, please feel free to contact us: [email protected]

The future intelligent system will increasingly rely on high-quality datasets to optimize decision-making and automated processes. In the era of data, companies and researchers need to continuously improve their ability of data collection and annotation to make sure the efficiency and accuracy of AI models. To gain an advantageous position in fiercely competitive market, we must laid a solid foundation in data.