We describe an embedded vision system, which integrates a web-cam quality CMOS imaging chip with a RISC processor, to perform real-time car-counting functions in the indoor and outdoor environment. The challenge of this application, especially for the outdoor environment, is to develop vision algorithms for day and night, and during the light-transition periods (i.e., dawn and dusk). The vision system also needs to accommodate a tremendous range of illumination change (from sunny summer to snowy winter). Finally we report briefly the result of an outdoor system we deployed in Germany since June 2003. The entire system consists of a network of 13 embedded vision systems covering a parking facility over 1 square kilometer. The vision-network has been in daily use for the employee parking guidance.