This report describes a real-time sensory system for monitoring safety in work zones by processing sequences of greyscale images from a stationary camera.
The system uses the difference of high- and low-frequency elements to accomplish the goal of detecting the moving object. The output of detection phase, the tracked regions (or blobs) are passed to the classification level, and to the safety monitoring level. The classification level deals with the models of moving objects in the work-zone, and it depends on the results of the detection stage as the only source of input. Thus, researchers use processed images rather than raw images, which are highly sensitive to noise.
In the safety monitoring stage, the system continuously detects situations, which potentially may cause accidents, and outputs warning messages if applicable; Designed to be robust, close to real-time, and easy to operate, the system was implemented on a Datacube Max Video 20 equipped with a Datacube Max 860 vector processor, and was converted later to new generation Datacube Max Video 200 with Max 860. Experimental results based on footage from several work-zone sites around the Twin Cities metro area have shown that the system is robust in different situations.
