Abstract: With the continuous development of automatic driving technology, the requirements for the accuracy of 3D target detection in complex traffic scenes are getting higher and higher. To solve the problems of low recognition rate, long detection time, and poor robustness of traditional detection methods, this paper proposes a new method based on PointFusion model improvement. The method utilizes the PointFusion network architecture to input 3D point cloud data and RGB image data into the PointNet++ and ResNeXt neural network structures, respectively, and adopts a dense fusion method to predict the spatial offsets of each input point to each vertex in the 3D selection box point by point, to output the 3D prediction box of the target. Experimental results on the KITTI dataset show that compared with the PointFusion network model, the improved PointFusion-based model proposed in this paper improves the 3D target detection accuracy in three different difficulty modes (easy, medium, and hard) and performs best in the medium difficulty mode. These findings highlight the potential of the method proposed in this paper to be applied in the field of autonomous driving, providing a reliable basis for navigating self-driving cars in complex environments.

Jun Wang, Shuai Jiang, Linglang Zeng and Ruiran Zhang, “Research on 3D Target Detection Algorithm Based on PointFusion Algorithm Improvement” International Journal of Advanced Computer Science and Applications(IJACSA), 14(11), 2023. http://dx.doi.org/10.14569/IJACSA.2023.0141139

@article{Wang2023,
title = {Research on 3D Target Detection Algorithm Based on PointFusion Algorithm Improvement},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2023.0141139},
url = {http://dx.doi.org/10.14569/IJACSA.2023.0141139},
year = {2023},
publisher = {The Science and Information Organization},
volume = {14},
number = {11},
author = {Jun Wang and Shuai Jiang and Linglang Zeng and Ruiran Zhang}
}