This work presents a vision-based obstacle detection and avoidance method for autonomous mobile robots. The implementation of an algorithm able to navigate a robot in arbitrary environments usually demands of the synergy of several sensors. This work presents an algorithm employing only one sensor, i.e. a stereo camera, thus significantly diminishing the system’s complexity. The implementation of this algorithm can be divided into two separate and independent modules. First, the stereo vision module retrieves information from the environment and produces disparity maps and then the decision making module analyses the data of the disparity maps and governs the robot’s direction. The achieved frame rate ensures that the robot will have enough time to accomplish the proposed decisions in real time. Both of the modules have been implemented in C++. The complete algorithm has been examined by being applied on an extensive set of pre-captured stereo images.