With the rapid development of AI technology, datasets has become a core factor of improving intelligent system’s performance. The variety and accuracy of datasets determine the learning ability and execution effect of AI models. In the progress of training intelligent system, large amount of datasets from real world are indispensable resources. Collecting and labeling data scientifically can help AI models gain accurate results in real applications, reduce the rate of misjudgment, and improve user experience and system efficiency.

Principle of Gait Recognition

The most common gait recognition system is based on four components: capturing gait data, analyzing pictures, feature extraction, and data comparison.

● Capturing gait data: The gait data is collected by cameras from different angles, and the video sequence of the gait is obtained through detection and tracking.

● Analyzing pictures: The gait characteristics are extracted through preprocessing and analysis. This includes the key processing of gait recognition, such as motion detection, motion segmentation, feature extraction, etc., of the gait movement in the image sequence.

● Feature extraction: Process the gait data to make it same mode as the gait stored in the database.

● Data comparison: Compare and identify the newly collected gait features with the gait features in the database.

Advantage of Gait Recognition

From a biological point of view, different people have different leg bone lengths, muscle strengths, center of gravity heights and nerve sensitivity, which determine the uniqueness and stability of gait, so it is difficult to be imitated by others in a short time.

From a hardware point of view, gait recognition doesn’t require high hardware requirements. Generally speaking, a 1080P camera has an effective recognition distance of up to 50 meters. If the resolution reaches 4K high-definition, the effective recognition distance can be extended to 100 meters. And it is full-view recognition, no matter what direction people come from, the gait can be recognized. What’s more, gait features can be collected and recognized without the cooperation of the recognized people.

Application of Gait Recognition

In the field of public security criminal investigation, gait recognition can quickly search for targets or video clips from a large number of videos in the situation of changing clothes, cross-scene, and face occlusion, through the analysis of the characteristics of target person’s height, posture, movement pattern, etc. This can make up for the shortcoming of face recognition technology.

In the field of smart home, gait recognition can be well applied to smart home systems, giving home appliances intelligent perception and providing more personalized services. At present, smart home systems on the market mostly use human body detection sensors or face recognition perception modules, but these methods have certain limitations. The human body sensor cannot intelligently distinguish whether it is an elderly person, a child or a pet, and face recognition needs the cooperation of recognized people, which basically cannot work in the night. The characteristics of gait recognition that are not sensitive to the lighting environment and do not require cooperation can make up for these application weakness and truly achieve personalized services.

As world’s leading AI data service provider, Nexdata developed “10,000 People — Simulation Monitoring View Gait Recognition Data” and “100 People — Gait Recognition Data”. Nexdata strictly abides by relevant regulations, and the data has been authorized by the person to be collected.

10,000 People — Simulation Monitoring View Gait Recognition Data

Gait recognition data for 10,000 people. 12 cameras were used to capture three walking rhythms: fast, medium, and slow. The data diversity includes multiple age groups, multiple time periods, 3 clothing states, and a total of 360,000 videos.

100 People — Gait Recognition Data

Gait recognition data for 100 people. Data scenes include indoor and outdoor scenes. The data diversity includes 10,498 videos in multiple age groups, multiple time periods, 5 scenes, 3 clothing states, and different collection angles. The accuracy of data labeling exceeeds 95%.

The future of AI is highly dependent on the support of data. With the development of technology and the expansion of application scenarios, high-quality datasets will become the key point to promoting AI performance. In this data-driven revolution, we will be able to better meet the opportunities and challenges of technology development if we constantly focus on data quality and strengthen data security management.