Due to the rapid increase in the amount of available video data, there has been a growing demand for efficient methods to understand and manage the data at the semantic level. In this paper, the V-OWL is proposed with extensions to OWL, which can describe complex video content including temporal-spatial and uncertain relationships. The B-Graph description model based on Bayesian Net is proposed to map the concepts and relationships in V-OWL ontology into the nodes and edges in B-Graph. Video semantic content can be discovered automatically by using existing training and reasoning methods of Bayesian Net. Results from experiments show that V-OWL has achieved good description of complex video content, and satisfactory precision and recall of high level semantic content detections.