An Improved Detection Algorithm For Pre-processing Problem Based On PointPillars

Jing Zhang, Da Xu, Jiajun Wang, Yunsong Li

Abstract: In the field of automatic driving, target detection by processing lidar point cloud is one of the key points. In order to effectively utilize point cloud information and reduce the information loss, this paper proposes a new pseudo-image generation method. Inspired by the basic idea of attention mechanism, a two-dimensional fusion attention module is proposed, which fuses features from multiple pseudo-images in height and channel dimensions, enhancing the ability of expression of feature information. The results show that the improved network performs better than the original network.