CORI: A configurable object recognition infrastructure

In this work, we present an object recognition infrastructure that can be adapted to the needs of a broad spectrum of tasks without writing a single line of code. Instead, it can be configured to generate various visual features and to perform training and recognition. The configuration format is quite simple and tools to generate these configurations make it easy to use even for developers with little computer vision experience. New algorithms and approaches can be added in a reusable way and further tools are provided to cover all parts of object recognition applications from data acquisition over execution and storage to the final result presentation. Moreover, performance considerations have been taken into account to enable the reuse of intermediate results and to support multiprocessor architectures. We demonstrate the use of the proposed infrastructure in a case study and mention tasks where it has already been used.