A method of measuring quality of algorithm in stream structure from Motion applied to aerial photography

Structure from Motion problem has many solutions nowadays. However, only a small number of them try to work in real-time restrictions, where reconstruction can be not perfect, but must be obtained in very short time. We introduce stream SfM problem, in which input is a stream images, and output us a stream of reconstructions corresponding to each time moment. To solve this problem, we start from developing a method that can measure results of stream SfM algorithms. It must pay equal attention to the quality of reconstruction and speed of its estimation. With focus on SfM applied to aerial photography, we propose such measuring method and use it to evaluate a bunch of algorithms, including one, which we develop specially for this quality measuring.