English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Obstacle detection is a safety-critical problem in robot navigation, wherestereo matching is a popular vision-based approach. While deep neural networkshave shown impressive results in computer vision, most of the previous obstacledetection works only leverage traditional stereo matching techniques to meetthe computational constraints for real-time feedback. This paper proposes acomputationally efficient method that employs a deep neural network to detectoccupancy from stereo images directly. Instead of learning the point cloudcorrespondence from the stereo data, our approach extracts the compact obstacledistribution based on volumetric representations. In addition, we prune thecomputation of safety irrelevant spaces in a coarse-to-fine manner based onoctrees generated by the decoder. As a result, we achieve real-time performanceon the onboard computer (NVIDIA Jetson TX2). Our approach detects obstaclesaccurately in the range of 32 meters and achieves better IoU (Intersection overUnion) and CD (Chamfer Distance) scores with only 2% of the computation cost ofthe state-of-the-art stereo model. Furthermore, we validate our method'srobustness and real-world feasibility through autonomous navigation experimentswith a real robot. Hence, our work contributes toward closing the gap betweenthe stereo-based system in robot perception and state-of-the-art stereo modelsin computer vision. To counter the scarcity of high-quality real-world indoorstereo datasets, we collect a 1.36 hours stereo dataset with a mobile robotwhich is used to fine-tune our model. The dataset, the code, and furtherdetails including additional visualizations are available athttps://lhy.xyz/stereovoxelnet

StereoVoxelNet: Real-Time Obstacle Detection Based on Occupancy Voxels from a Stereo Camera Using Deep Neural Networks