Research on 3D Object Detection in Intelligent Construction Based on Computer Vision

To solve the problem of large scale difference and unsmooth multi-scale feature fusion in monocular 3D target detection, a multi-scale feature fusion pyramid network is proposed in this chapter. The network includes the backbone network and the multi-scale pyramid network. The backbone network extracts the feature information of different scales, and integrates the shallow feature information and deep feature information of different scales with the buffer module to solve the problem of uneven feature fusion. Multi-scale pyramid network fully fuses different scale feature information through multiple up-sampling features, and uses attention mechanism to reduce the interference of useless information on subsequent feature fusion. Finally, experimental results on KITTI dataset show that compared with other previous monocular 3D object detection methods based on multi-scale feature fusion, the average accuracy of 3D enveloping frame and BEV in this chapter is optimal at simple, medium and difficult detection levels.