Improved Intersection Multi-Source Sensing Fusion Method of YOLOX

In the intersection scene, millimeter wave radar has poor perception effect on static target, and its visual perception has the problem of missing detection of small target. In order to solve these problems, the convolutional attention mechanism is introduced in this paper, and VariFocal loss and CIoU loss are used to optimize the loss function of YOLOX to improve the detection performance of small targets. To solve the problem of static target missing detection, the sensor fusion algorithm of millimeter wave radar and camera based on double similarity index is designed, and the sensor fusion accuracy is improved by the target state correction method. Finally, the algorithm is transplanted to the roadside edge computing equipment, and the test data is collected and analyzed under different weather conditions. The results show that the accuracy of perception fusion algorithm can reach 91.33% in sunny, cloudy and night scenarios.