Learning to Remove Multipath Distortions in Time-of-Flight Range Images for a Robotic Arm Setup

摘要

Range images captured by Time-of-Flight (ToF) cameras are corrupted withmultipath distortions due to interaction between modulated light signals andscenes. The interaction is often complicated, which makes a model-basedsolution elusive. We propose a learning-based approach for removing themultipath distortions for a ToF camera in a robotic arm setup. Our approach isbased on deep learning. We use the robotic arm to automatically collect a largeamount of ToF range images containing various multipath distortions. Thetraining images are automatically labeled by leveraging a high precisionstructured light sensor available only in the training time. In the test time,we apply the learned model to remove the multipath distortions. This allows ourrobotic arm setup to enjoy the speed and compact form of the ToF camera withoutcompromising with its range measurement errors. We conduct extensiveexperimental validations and compare the proposed method to several baselinealgorithms. The experiment results show that our method achieves 55% errorreduction in range estimation and largely outperforms the baseline algorithms.