Abstract: A novel algorithm to detect objects across multiple views at the same instance is proposed based on object probabilistic random field (OOPF). OOPF is the core of our algorithm and it integrates all the information of individual views into the fusion space. Then object detection is implemented in the fusion space. An object window structure is designed in order to compute OOPF fast, which also provides a scheme to detect objects and the correspondence simultaneously. The experiment results demonstrate that the algorithm has a better performance against occlusion while the computational complexity is low.