Dynamic speckles are produced when a diffuse object moving along x-axis with constant velocity is illuminated by coherent light. The diffuse object is considered as a weak random phase screen, statistical properties of dynamic speckles in the Fresnel diffraction field are given in this paper. It is shown that the statistical properties of dynamic speckles depend on the optical condition, surface roughness, and moving velocity of the diffuse object. The parameters describing the property of dynamic speckles extracted from the speckle image, such as fractal dimension, speckle contrast, gray feature of speckle image, binary feature and speckle size. Then according to these features parameters, the recognizing system is established based on neural network technique. Four metallic flat-grinding samples, moving with velocity 4mm/s and 16mm/s respectively, are measured the roughness using this experiment set-up and recognizing system. It’s shown that the average roughness data Ra obtained using the new technique and the Ra obtained using Talysurf are in good agreement. The experimental set-up of this method is simple, fast, and not sensitive to change of circumstance and vibration. Hence, it has great potential for application to real-time measurement.