Crowd behavior patterns recognition and age/gender estimation
Problem and task description
Nowadays business can easily get access to lots of useful information about human behavior patterns from security and other types of cameras. This information can be used for marketing purposes or for security improvements. AI can quickly extract this info and convert it to visually appealing reports, or provide important alerts.
Our client wanted to use the latest technologies in the field of computer vision to create video-analytical service for deeper customer behavior understanding in their shopping mall. This service had to provide with the answers to following questions:
- What are the most popular routes people take while walking around the shopping mall?
- What are the most popular areas of the shopping mall?
- What gender and age proportion at what time enters the shopping mall?
- Is there anybody who might be considered as dangerous?
- How many visitors are in a certain area at a given time point.
Our solution
- We’ve developed a service, capable of processing huge amounts of data from a significant number of cameras, providing a real-time people detection, tracking and re-identification.
- We’ve developed and successfully implemented the original method for age/gender estimation on low quality videos.
- For the task of people counting we’ve used a combination of several crowd counting methods.
