To address the limitations of the PointPillars algorithm, including insufficient accuracy and issues with missed detections and false alarms, an enhanced 3D small object detection algorithm based on PointPillars is proposed： A lightweight attention mechanism, ECA, is integrated into the backbone network to enhance the network′s feature representation capabilities, thereby improving detection accuracy. Additionally, to achieve better performance in deep models, we replace the ReLU activation function with the Hardswish activation function, effectively mitigating the gradient vanishing problem. This improved framework is validated on the KITTI dataset. Experimental results demonstrate a substantial enhancement in the algorithm′s ability to detect small objects. Under the bird′s-eye view, the average mean average precision (mAP) for medium difficulty targets (MmAP) reaches 71.31%, representing a 5.62 percentage point improvement compared to the original algorithm. Furthermore, the proposed model exhibits superior performance in detecting small and occluded objects compared to other mainstream models.