You are here

Ir. Maurice Peemen

PhD candidate Faculteit Electrical Engineering, Parallel Architecture Research Eindhoven

Developing a biologically inspired trainable computer vision platform. This involves training algorithms that also perform desing space exploration for a given vision task. Develop theories and methods for application optimization by code transformations for data locality. Development of a hardware architecture which is parameterizable for different application demands.

My work for COMMIT/

The overall challenge of the SenSafety project is to offer real-time automatic analysis of potential hazardous situations. This includes detection of important events, and to give support in such situations to the first responders. An example is video surveillance, which can be used to track specific objects in a scene. An important focus of the SenSafety project is the image processing that is required to perform detection and recognition tasks. This processing is performed locally close to the sensor (e.g. Smart Camera), such that privacy in the area of the sensor can be protected. Processing the video streams locally is a major challenge for the current generation of restricted embedded platforms. Furthermore, these platforms should be flexible such that different detection targets can be easily select at run-time.

To cope with these requirements, I study biologically inspired classification algorithms that can be reprogrammed by machine learning, (e.g. Convolutional Neural Networks). To perform detection tasks with real-time constraints, I develop dedicated accelerator architectures, that achieve excellent performance by exploiting extreme levels of data locality to minimize data transfer. Eventually, this leads to accurate but flexible detection platforms with an excellent energy-efficiency.

Works with