Video Analysis

Computer vision is shaping and driving research and development in the fields of machine learning and artificial intelligence. Especially Facial analysis in plays a central role here.

The face is the most important external characteristic of a person and consists of a unique combination of individual features. These features are not only used to distinguish known from unknown people. They also help to assess a person's age or emotional state, for example.

Facial analysis is a branch of image and video analysis in the field of computer vision, which deals with the computer-assisted processing and analysis of camera images. It involves the automatic extraction of a wide range of information from image and video recordings of faces. This procedure enables the extraction of features that can be used to detect or recognize people, but also to differentiate between people by age or perceived gender, for example.

The amount of data generated by camera recordings is extremely large compared to other sensor data, especially when it comes to video recordings.. This results in challenges for the entire processing chain, which are reflected in the complexity of the networks used, the size of the models and the computing power required, among other things. Compared to the processing of individual images, these requirements increase even more dramatically when analyzing videos (moving images).

This results in various challenges and questions:

• How can methods and processes be transferred from still images to moving images?
• How can existing facial models and analysis processes be further optimized?
• How can the AI/ML models be scientifically evaluated and corrected in order to avoid unintended bias in the results?
• How can the accuracy of perceived gender and age classifications be further improved?
• How can analysis processes also be implemented on small devices (SoC)?

In addition to the technical challenges, especially facial analysis also raises questions about social acceptance (privacy and data protection) as well as fairness and trustworthiness (bias).