Video surveillance has developed considerably in the recent years. Analyzing the data generated by such systems has become a major challenge. To address this issue, we propose a framework for the creation of rich Labeled Property Graphs from video surveillance streams. It is based on 1) a Deep Learning pipeline architecture for video data extraction, 2) a graph database module to efficiently structure and store detections, and 3) a querying module to interact with generated graphs, enhancing the automatic analysis of scenes. Its modular architecture enables the feature extraction steps from the videos to be easily maintained, modified or interchanged. Our demonstration scenario shows the process of generating scene graphs from videos of several benchmark datasets. The audience will assist to an end-to-end execution of the pipeline showing the generation process and visualize generated graphs. They will have the opportunity to formulate queries using an interface illustrating several use case scenarios involving person re-identification and abandoned objects matching with their former owners.