Object recognition and simultaneous indoor location algorithm with stereo camera

Visual SLAM mainly depends on the measurement data of camera for dead reckoning, it is difficult to maintain the positioning accuracy and stability for a long time. Therefore, visual SLAM can't be widely used in indoor positioning at present. The algorithm of object recognition and simultaneous indoor location is proposed in this paper, which combines the technology of object recognition based on deep learning, image feature matching and binocular parallax positioning. Using a low cost binocular stereo camera to output the object category and its three-dimensional position in camera coordinate system. Finally, the static and dynamic indoor positioning tests are carried out, the indoor horizontal positioning error for static test is 74 mm, the indoor horizontal positioning error for dynamic test is less than 250 mm. When the distance between object and camera becomes larger, the positioning error becomes larger. When the parallax becomes larger, the positioning error becomes smaller.