<p>Monocular 3D object detection has gained extensive attention due to its low cost and ease of deployment, yet remains challenged by missing depth cues and occlusions. To address these challenges, we propose a novel framework named cross-scale feature enhanced voxel network (CSFEVoxNet). The method introduces the cross-scale feature enhancement (CSFE) module, which employs an innovative lightweight convolutional structure with a global–local feature aggregation strategy for efficient multi-scale feature extraction and fusion. Particularly in occlusion scenarios, this module leverages the complementarity between global features and local details to significantly improve detection performance for partially occluded objects. Moreover, the CSFEVoxNet method lifts the 2D feature map into a structured voxel grid by using the camera’s intrinsic and extrinsic parameters and depth information, constructing feature voxels containing the 3D structural information of objects. This computationally intensive approach provides richer geometric context for object detection tasks and significantly improves the model’s ability to perceive the position, size, and shape of objects. Experiments demonstrate that the proposed method outperforms existing state-of-the-art approaches on large-scale published indoor 3D object detection datasets, especially showing advantages in handling small objects and complex occlusion scenarios.</p>