A method for detecting obstacles in a field scene based on depth values and semantic probabilities

Abstract

The application discloses a kind of field scene obstacle detection method based on depth value and semantic probability, for the key difficult problem of unstable single mode perception in the complex natural environment of field, with the difference of significant and significant difference of single mode perception, the detection scheme of a kind of fusion depth gradient and semantic probability is proposed, with the consideration of environmental adaptability, detection robustness and accuracy.The depth gradient analysis is used, the depth mutation area is extracted by image preprocessing and dynamic threshold strategy, and the significant boundary of scene space structure change is accurately described.The semantic probability candidate region generation is used, the joint screening is carried out in combination with semantic confidence and depth significant boundary, the single channel error is effectively inhibited, and the detection integrity is guaranteed.The screening result is clustered, merged and denoised using the density-based spatial clustering and boundary refinement method, and the structure complete and spatial coherent obstacle boundary box is obtained, with strong anti-interference ability and high detection precision.

Images