3D Object detection from multimodal data, mainly camera and LiDAR sensors, has been studied extensively, especially in the wake of fusion algorithms. LiDARs can procure accurate depth information for precise localization. However, they suffer from being sparse generally in faraway objects, resulting in reduced recall values. Many algorithms have tried to improve the detection recall by enhancing the point density in the sparse areas. In this paper, we try to improve the point cloud sparsity using pseudo points generated from the corresponding 2D images procured by the camera sensor. We start by extracting 2D bounding boxes from the RGB image and creating 3D pseudo points from the 2D pixels within the detected regions. The pseudo points are combined with the actual LiDAR points which are then subjected to a LiDAR 3D detection pipeline based on Voxel-RCNN. The model is trained and evaluated on the KITTI 3D detection benchmarks showing that it performs significantly faster (50 ms) than the state-of-the-art methods at comparable detection accuracy.