Video & Image Analysis

The HD-RNN framework is optimally suited for applications involving analysis of video streams and still images in real-time. The technology is effectively deployed on both ends of the performance spectrum:

•  Low-cost, resource-constrained embedded environments
•  Mobile devices
•  Video surveillance
•  Service robots
•  Accelerated-performance server solutions which handle large amounts of data in real time

Unlike existing solutions, there is no need for application-specific, rule-based programing. The technology utilizes a self-learning mechanism for finding patterns and objects in 2D, 3D, and N-dimensional data. Fusion of data originating from different sensory modalities (e.g. visual, audio, etc.) is intrinsically supported.

A single GPU card running HD-RNN server suite handles more than 500 distinct object recognitions per second. For example, a server with 4 GPU cards can easily support over 1,000 recognitions per second, or several live video feeds.

Examples of applications to which HD-RNN was successfully applied to include:

•  Letter/number recognition
•  Emotion identification
•  Face recognition
•  Object identification [video]
•  Behavior detection [video]