Measuring the exposed gravel surface geometry is important for understanding the flow characteristic. This research has developed a reliable and precision methodology to generate the Digital Surface Model (DSM) of gravel surfaces using multi-scale images from close-range photogrammetry. The camera was calibrated on-site corresponding to different scale images with a self-calibration bundle adjustment using coded retro-reflective targets and the DSM of the gravel surface was reconstructed for two scales from different scale images relatively. The small scale DSM was used as the control surface covering entire area (4.5 m × 4.5 m). The large scale DSMs contain more detailed elevation variation covering the diagonals (6 m × 1 m) of entire area and are registered to the small scale DSM by iterative closest point (ICP). Three fields with a size of 4.5 m × 4.5 m have been tested and discussed, and the results are provided and validated with high accuracy total station. The ortho-images can be generated and apply for determination of grain size distribution.