Programmable logic enables single-chip target tracking
Andrew Wilson
Video-motion tracking systems are used in many imaging applications, including intelligent traffic control and security. Current approaches to motion tracking include slow software-based systems and expensive hardware-based embedded systems that use pipelined processors and multiple frame buffers. In both types of systems, motion tracking transforms objects in the video image into object coordinates.
"Connectivity analysis is the bottleneck for video-processing functions that convert video data to pixel groups," explains Craig Sullender ([email protected]), president of MoTarg (Austin, TX). "But once a target is identified, high-bandwidth data analysis no longer needs to be performed." To perform connectivity analysis on video data, MoTarg has developed an object-targeting algorithm that is stored in a programmable logic device (PLD) from Altera (San Jose, CA). The algorithm uses a single set of minimum and maximum coordinates to mark the extent of an object. For object shape definition, an object`s target can be broken up into several smaller targets.
"These coordinates provide an easily scalable representation of shape that are well suited to pattern matching," says Sullender. "Other information, such as object color, can be part of the target information. By marking different portions of a person`s body with color tags, for example, separating and following the hands and feet becomes straightforward," he explains.
At the Vision Systems Design Intelligent Vision `99 conference (Santa Clara, CA, June 28-29; see Vision Systems Design, Aug. 1999, p. 19), MoTarg demonstrated a stand-alone video-evaluation board for motion and color targeting using the Altera EPF10K100-3 PLD. By applying connectivity analysis directly to the video stream, the board eliminates frame buffers. In operation, it produces up to 254 targets in every field. Object area and object center can be included in the output list of targets. After targets are acquired, the objects can be tracked by following their paths over time.
MoTarg is offering its targeting technology for licensing as a Verilog core, a PLD, or a full-custom chip called the Video Target ASIC (VideoTA). Related technologies available for integration with the VideoTA include serial ports, motion tracking, and object recognition firmware.
According to Sullender, existing products can gain increased functionality from inexpensive targeting. In manufacturing, for example, object recognition and pattern matching are used for parts alignment, sorting, and inspection. Instead of searching the complete video scene for objects, vision systems can use targeting for feature extraction. Only the sections of the video scene containing useful information are passed to the recognition system.
"Intelligent video cameras with adjustable windows can use targets to size and move the windows to follow objects of interest," says Sullender. "By reducing the amount of data from CCD cameras, target windows can respond in real time to object movements." Other applications that might benefit from inexpensive target tracking include security, motion analysis, and robotic vision systems.
