| Περίληψη: | On-line monitoring is an important challenge in future biotechnology applications, for instance in the domain of precision livestock farming, there is need for low-cost intelligent sensors to monitor animal welfare. The common way of observing a living organism is usually done by audio-visual ways performed by a human being, who is present on the scene. This method is, however, subjective, expensive, error prone and time consuming. Instead of performing an animal observation by a human being, automated objective surveillance, by means of low cost intelligent image sensors, can be used. With the use of cheap image sensors and with the help of image analysis techniques, an automated, objective, contact-less monitoring method of the behavior of the living organisms can be provided.
Much knowledge has been obtained in the development and use of image analysis algorithms to automatically quantify body features of animals, their activity rate and their behavior. Such an automatic image analysis algorithm is combined with on-line modeling techniques in order to develop an application for the recognition of several behavioral phenotypes of laying hens. The procedure is divided in two phases, where an automatic computer vision algorithm detects the monitoring object from images captured by a video camera, and then another algorithm tracks the detected object through successive frames.
Further work is required to integrate these algorithms into low-cost low-energy processing platforms, including embedded systems or even wearable devices. Only then, this important biotechnology development will lead to economically applicable solutions. The challenge of the present thesis especially includes the exploration of ultra-low energy implementation platforms of this biotechnology application. The initial application is developed in the MATLAB environment and is converted to C programming language. Dynamic range and precision analysis are performed to efficiently determine the required fixed-point word-lengths of the application’s variables. Finally, platform-independent and platform-dependent code transformations and integration of the algorithm to different ASIPs (Application Specific Instruction Processors) architectures are applied in order to achieve ultimate low energy consumption.
|
|---|
