<p>3D object detection has attracted growing research attention due to its expanding range of applications in autonomous driving perception. However, existing detection algorithms are prone to miss detection and false detection for long-range small targets, occluded targets, and targets with similar appearances in complex traffic scenarios. To address these challenges, we propose a novel 3D object detection algorithm based on the PointPillars framework. Specifically, we first design a point cloud feature extractor based on a triple attention mechanism to enhance target features and suppress background noise. Furthermore, we construct a new backbone network using RepViT Blocks, which decouples token and channel mixers and integrates structural reparameterization to balance model performance and computational efficiency. Finally, we develop a multi-scale dual detection head that performs predictions on feature maps of different resolutions, thereby improving the recall rate in complex scenarios. Experiments on the KITTI dataset demonstrate that the proposed model achieves 3D mAP of 81.67%, 60.05%, and 71.05% for cars, pedestrians, and cyclists, respectively, achieving improvements of 2.28%, 11.29%, and 9.82% compared to the baseline model. These results validate the effectiveness of our method.</p>