Understanding the impacts of climate change on plant phenology is crucial for predicting ecosystem responses. However, accurately tracking the flowering phenology of individual plant species in grassland species mixtures is challenging, hindering our ability to study the impacts of biotic and abiotic factors on plant reproduction and plant-pollinator interactions. Here, we present a workflow for extracting flowering phenology from grassland species mixtures using near-surface time-lapse cameras. We used 89 image series acquired in plots with known species composition at the Jena trait-based experiment (Germany) to develop random forest classifiers, which were used to classify images and compute time series of flower cover for each species. The high temporal resolution of time-lapse cameras allowed to select images in proper light conditions, and to extract vegetation indices and texture metrics to improve discrimination among flowering species. The random forest classifiers showed a high accuracy in predicting the cover of Leucanthemum vulgare, Ranunculus acris, and Knautia arvensis flowers, whereas graminoid flowers were harder to predict due to their green-to-brownish colours. The proposed workflow can be applied in climate change studies, ecosystem functioning, plant community ecology, and biodiversity change research, including the investigation of effects of species richness on individual species' flowering phenology. Our method could be a valuable tool for understanding the impacts of climate change on plant reproduction and ecosystem dynamics.