An unmanned aerial vehicle (UAV) may select a subject to track or follow and capture images including the subject. A visual recognition system of the UAV may generate a profile of reference images portraying the subject breaking the reference images down into pixel sets corresponding to the subject and his/her components and distinguishing characteristics (body parts, clothing, facial features, accessories). The visual recognition system may break the incoming stream of incoming images into pixel sets, analyzing the pixel sets to distinguish the subject from his/her surroundings (i.e., in a crowd) and determine movement of the subject and the current orientation of the UAV to the subject. The UAV may then change its heading, velocity, or position based on any difference between the current orientation and the desired or predetermined orientation between the UAV and the subject.