Using Stereo Vision to Pursue Moving Agents with a Mobile Robot

CAIR-2 is an intelligent mobile robot developed for guidance and delivery. One of the major features of CAIR-2 is that it provides the active vision capability based on real-time visual feedback using on-board low-cost processors. Instead of using the expensive special image processors, we exploit conventional processors such as Motolora 68040 and reduce processing time using the smart system and vision software. Vision system of CAIR-2 consists of real-time kernel, image saver, database, and two vision modules. First Stage Vision Module (FSVM) segments 256x256 gray-scale images into a set of regions using the partition-mode-test algorithm. Targets are then recognized by their shapes. Once targets are initially located by FSVM, Second Stage Vision Module (SSVM) can easily find and keep tracking targets using the focus-of-attention strategy based on Kalman filter. In case false targets are detected, the verification module can rapidly correct them. Real-time kernel, image saver, database are being exploited to manage and process the large image data efficiently and thus reduce the overall processing time. Combining the above four mechanisms effectively, while robot moves around both indoors and outdoors, vision system of CAIR-2 can recognize and track multiple moving targets simultaneously every one thirtieth of a second in average.