A 3D Object Detection Algorithm Based on Improved EPNet

Early 3D object detection algorithms focused on using a single sensor. However, with the development of target detection technology, the target detection algorithm based on multi-sensor fusion has gradually entered people's field of vision. In this paper, we improve the EPNet algorithm based on image and point cloud fusion. In its image stream branch, due to its relatively simple image upsampling method, most of the image information will be lost, which reduces the detection accuracy. Therefore, the method of combining the feature pyramid network with the coordinate attention mechanism is used to make up for the problem. Extensive experiments on the KITTI datasets demonstrate this method is good.