Although current autonomous driving perception systems have made significant progress, research on the widely used six-views visual setup in practical applications remains limited. Addressing the need for training and testing autonomous driving models, existing virtual datasets predominantly use four-camera inputs and do not support simultaneous lane and object detection. While real-world datasets are more realistic than virtual ones, they fail to provide a safe and controlled testing environment for autonomous driving systems, and they incur high testing costs. To address these issues, this study constructs a virtual 3D detection dataset for the multiple visual sensors fusion of autonomous driving(MultiVir). To demonstrate the effectiveness of the dataset, we establish an autonomous driving system capable of simultaneous object and lane detection. The results indicate that the layout of the six views imposes stricter requirements on model performance.