In this study, we evaluate the influence of different cloud point filtering algorithms on the process of accurately dimensioning objects. This is critical in vision-based measurement systems, particularly for logistics and packaging applications. We assess three smoothing algorithms: bilateral filtering, statistical outlier removal, shadow filtering algorithm alongside baseline unfiltered data. We extract object dimensions by fitting a convex hull applied to the processed point cloud, and evaluate across different positions, parcel types, and edge lengths. We employ various statistical metrics to evaluate algorithm performance. Our research utilizes point clouds of cardboard boxes for evaluation, collected with the ToF Kinect v2 depth camera. Study includes both cuboidal objects and distortion-simulated shapes. We assessed a dataset of 639-point cloud samples. The data was collected under controlled lighting with top-down camera orientation and processed using the PCL library. Our findings show that shadow filtering consistently and significantly outperforms the other methods on standard cuboid geometries. However, in the presence of shape distortions, it occasionally introduces large-magnitude outliers, reflecting overly aggressive filtering behaviour. Additionally, we observe scale-dependent error pattern across all object types, with dimensional accuracy decreasing as object size increases.