Free space optical (FSO) communication systems have a myriad of applications, from secure high-speed links to use in radio frequency (RF) challenged or restricted environments. Because FSO communication systems have improved performance when the transceivers are well aligned, and since many practical applications are nonstationary, low-latency acquisition, tracking and pointing (ATP) is crucial. In this paper, we will present an ATP system based on computer vision, suitable for various scenarios including robotics, vehicular communications, and indoor optical wireless. For this system, a camera is integrated with a gimbal-less two-axis micro-mirror array driven by a micro-electro-mechanical system (MEMS) on the transmitter (base station) side. The receiver (mobile device) is marked by a landmark, so the direction and relative position from the transmitter to the receiver can be estimated by applying image recognition and computer vision techniques (note that the latency of the MEMS beam steering is low compared to the execution time of these image recognition algorithms). Once the receiver is acquired, optical flow algorithms (which are much less complicated than image recognition) are used to track the recognized landmark in real-time. When the camera and micro-mirror array are well calibrated, the transmitter is able to track the landmarked receiver in a room-sized space. We also analyze the trade-off among field-of-view (FOV), latency and reliability in this paper. Finally, we present FSO communication system link performance (e.g., latency, signal-to-noise ratio, bit error rate, reliability) between moving platforms incorporating the new ATP capability.