Software environment eases multiprocessing tasks
Partitioning image-processing algorithms across multiple processors is often a necessity in medical and military applications. Once code is written, it must be handcrafted to run on multiple processors. This process can now be automated by using software-development tools such as Rippen from Orincon Technologies (San Diego, CA).
The software`s graphical user interface lets designers partition image-processing code, resulting in a five to ten times improvement in system productivity, according to Byron Chen, Orincon`s development engineer. Multiprocessors from Mercury Computer Systems (Chelmsford, MA), Sky Computers (Chelmsford, MA), CSPI (Billerica, MA), and Mizar (Carrolton, TX) are currently supported under Rippen. Already, the US Army`s RDE Center (Ft. Belvoir, VA) is using Rippen as the basis of an intruder-detection system.
In operation, an infrared sensor is mounted on a robot to find intruders in a warehouse. Infrared images are processed at 10 frames/s by a multiprocessor architecture containing a 2CE host processor from Force Computers (San Jose, CA) and four i860 external processors. Acquired by a VME frame grabber bus, the images are modeled as a time sequence of random field, where the spatial mean of each frame varies with the baseline temperature.
Adaptive thresholding segments the possible human candidates, and moment invariants describe the statistical characteristics of the clusters obtained by segmentation. A neural network is applied to determine whether the cluster is a person or merely a clutter. The location of the person is then sent to the robot to pan the infrared sensor to track the human target.