We present a method to analyze images of the coma of 67P/Churyumov-Gerasimenko obtained using OSIRIS, the main imaging system onboard Rosetta, where dust aggregates can be seen as bright tracks because of their relative velocity with respect to the spacecraft. We applied this method to 105 images taken in 2015 July, 2015 December and 2016 January, identifying more than 20000 individual objects. We performed a photometric analysis of them, finding their phase function. This phase function follows the same trend as the one found for the nucleus, consistent with the detected particles having a size larger than ∼ 1 mm. Additionally, the phase function becomes shallower for increasing heliocentric distances, indicating a decrease in the mean agglomerate size. In order to characterize the agglomerates observed in the image, we developed a simplified model for their ejection and dynamics in the coma, and generated synthetic images based on it. We solved the inverse problem by finding the simulation parameters that give the best fit between synthetic and real images. In doing so, we were able to obtain a mean agglomerate size ∼ dm and initial speed 1 m s -1 . Both show a decrease with increasing heliocentric distance, sign of the reduction in activity. Also, the sizes obtained by the comparison are not compatible with ejection caused by water activity, so other sources have to be invoked, mainly CO 2 .