Abstract: Video big data has played an important role in the safety and production management of intelligent mines such as track safety, and has an important potential application in the dynamic monitoring of surrounding rock cracks. However, further research is still needed. The key lies in the extraction of crack information based on digital image time series. In the paper, visible-light images monitoring experiments on the rocks loaded in the process of failure were carried out. The light reflectivity response to rock damage and its spatial differentiation characteristics were researched. And a dynamic extraction method of cracks on rock stressed was proposed. In the method, time series of digital images characterized by light reflectivity were processed by calculus of differences, and the quantitative indicator, differential coefficient (C) who measures the space differentiation degree of outliers, was introduced. The results show that the growth and expansion of cracks on rocks stressed can cause the abrupt changes of light reflectivity, and the abrupt change rate can reach 0.2/s, which is much higher than that caused by other random factors (about 0.03/s). When the cracks are active, the abrupt changes of reflectivity show a significant spatial differentiation (C up to 189), and the degree of differentiation is much greater than that of random distribution (C = 1). The proposed cracks extraction method of rocks stressed based on the light reflectivity mutation and the spatial differentiation can be used to automatically judge the existence of cracks in the image and extract the instantaneous and accumulated information about crack activities, to study the damage evolution characteristics of rock (rock mass) stressed, and to understand the relationship between surface and internal damage of rock stressed. It can also provide experimental and method for references for the dynamic monitoring of surrounding rock fracture based on video big data of intelligent mines.