Advanced driver-assistance systems (ADAS) and autonomous driving depend on precise vehicle identification and localization. This study introduces a cross-modal framework that uses 2D image labels from a co-located and calibrated camera to estimate 3D orientated bounding boxes in LiDAR point clouds. 2D bounding boxes are projected into 3D frustums that restrict the search region in the point cloud using camera intrinsic and camera-to-LiDAR extrinsic parameters. Ground plane removal and Euclidean clustering are used to separate item candidates within these frustums, and 3D cuboid fitting is then used to estimate the bounding box precisely. Through the association of projected 3D LiDAR points with corresponding 2D image regions, the framework additionally calculates the distance between the ego vehicle and identified objects. The suggested approach improves perception pipelines through effective multi-sensor data fusion and consistently performs well in vehicle localization.