Of primary importance for mobile robots is their smooth movement to the targeted destination. To achieve this purpose, mobile robots must be able to detect a person in their environment, another mobile robot, or an object not described in the map and to avoid collision with it. Recognizing the strong need for providing robots with a visual system to evade obstacles, the authors first developed a real-time visual system to detect a moving obstacle and then studied the possibility of avoiding collisions by mounting the system on a mobile robot. The visual sensor used in this system is a passive optical stereo without any mechanical moving parts. Using a special slit patten, the sensor is configured in order to split the two images obtained by individual cameras place on the right and left and to project the split images onto one CCD sensor, providing approximately 200 auto-focusing subsystems. The sub-systems can operate independently of one another, enabling real-time processing. This paper reports on a visual sensor, a solution to the measurement accuracy problem concerning the detection of moving obstacles using the sensor, and visual system experiments on real-time detection of an actually moving object using the sensor.