Application of VR Driving Game in Visual Research of Unmanned Car

At present, most of the test of unmanned vehicles at home and abroad uses the construction of actual test sites to test the unmanned vehicles. However, the road conditions required for unmanned vehicle tests are relatively high, requiring a large amount of capital and labor. The material resources are very huge, and there are some natural environmental factors beyond human control. In view of the limitations of the above tests, it is of great significance to construct a safe, effective, economical and stable unmanned vehicle test environment. This article mainly studies the application of VR driving games in the visual research of driverless cars. The test environment of this paper includes virtual reality scene module, micro traffic simulation module, virtual reality interaction module, network communication module, and unmanned vehicle module. The unmanned vehicle on the test bench makes corresponding behavior decisions based on the received virtual scene data, and transmits these behavior decision data to the virtual unmanned vehicle in the virtual scene through network communication. The virtual unmanned vehicle operates in a virtual scene after obtaining data such as vehicle speed, heading angle, and brakes. This operating state is used to express the state of the real unmanned vehicle, thereby verifying the intelligence of the real unmanned vehicle. The experiments in this paper show that the network model has a high average accuracy rate under the same test time, and the detection speed can reach 13 frames per second. It can be seen that the overall performance of YOLOv3 is better. The test results show that YOLOv3 algorithm has better accuracy and real-time performance on seven common traffic sign detections.