Context Recognition Network (CRN) Toolbox

Description

Developing systems for real-time context recognition always implies building a streaming network for the sensor data. The data has to be continuously acquired from one or more sensors and then handed into diverse filters for pre-processing and feature extraction. The filtered data stream is then fed into the classifier for the context recognition. The classifier itself is often a nested structure or network of well known classification algorithms (HMM, Bayesian classifier, etc.). Most of these tasks are common in every recognition system and there is little sense in re-implementing them for each system.

The CRN Toolbox aims at a general solution for the streaming network behind real-time context recognition systems. It allows composing recognition systems from configurable components. The components are algorithms for acquisition, pre-processing (feature extraction), and classification. Algorithms may run as tasks on different hosts to cope with limited power of wearable computing environments but also to allow sharing of context information between cooperating systems. The tasks are connected through communication sockets, building a network for processing real-time sensor data.

The main features of the CRN Toolbox are:

• Easy and flexible composition of algorithms
• Tasks may execute on different hosts/operating-systems
• Possibility to integrate existing tools via Sockets.
• Easy extensibility

Use in wearIT@work project

Healthcare (Gespag)
Production (Skoda)
Maintenance (EADS) under investigation

Status

The toolbox is in use for the demonstrators, and under further development.

Maturity Level (?)