Embedded low-level video processing for surveillance purposes

This paper introduces an embedded architecture and the low-level video processing algorithms developed for an intelligent node that is a part of a distributed intelligent sensory network for surveillance purposes. In this paper, details of the architecture developed for this node are given, together with the low-level video processing algorithms used, as well as the results obtained after their implementation. The video board has been developed using two DSP processors for video processing tasks, as well as a FPGA dedicated to image capture (VGA size) and to dispatch them to the DSP processors. The low-level software includes acquisition, segmentation, labeling, tracking and classification of detected objects into three main categories: Person, Group and Luggage. Also, additional features are extracted from each object in the frame. The unit has to communicate the classification results and the main features obtained using XML streaming to upper levels, as well as the processed frames, using a JPEG stream. All these functionalities are currently running in the built prototypes.