With the trend of deep space exploration, visual based navigation, which contains high-precision space object reconstruction and spacecraft motion estimation functions, has become one of the key enabling technologies. An image-based space object reconstruction algorithm and a spacecraft motion estimation method are investigated in this paper. An image acquisition experiment testbed is developed to simulate the image acquisition process of space object in deep space. The relative position and attitude between images (or cameras) are obtained by spare bundle adjustment without preemptive knowing accurate spacecraft motion data. Furthermore, a preprocessing feature matching based incremental structure from motion algorithm is utilized to solve the problem of relative high portion of mismatch and removing the images with deficiency information. The algorithm is tested on the large photo collection. The effectiveness of algorithm is verified by error analysis between real position and estimated position.