eduzhai > Applied Sciences > Engineering >

MSDPN Monocular Depth Prediction with Partial Laser Observation using Multi-stage Neural Networks

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: In this study, a deep-learning-based multi-stage network architecture calledMulti-Stage Depth Prediction Network (MSDPN) is proposed to predict a densedepth map using a 2D LiDAR and a monocular camera. Our proposed networkconsists of a multi-stage encoder-decoder architecture and Cross Stage FeatureAggregation (CSFA). The proposed multi-stage encoder-decoder architecturealleviates the partial observation problem caused by the characteristics of a2D LiDAR, and CSFA prevents the multi-stage network from diluting the featuresand allows the network to learn the inter-spatial relationship between featuresbetter. Previous works use sub-sampled data from the ground truth as an inputrather than actual 2D LiDAR data. In contrast, our approach trains the modeland conducts experiments with a physically-collected 2D LiDAR dataset. To thisend, we acquired our own dataset called KAIST RGBD-scan dataset and validatedthe effectiveness and the robustness of MSDPN under realistic conditions. Asverified experimentally, our network yields promising performance againststate-of-the-art methods. Additionally, we analyzed the performance ofdifferent input methods and confirmed that the reference depth map is robust inuntrained scenarios.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×