New Foggy Object Detecting Model

TL;DR

This work tackles object detection under foggy, reduced visibility, a setting where accuracy and speed are challenged. It introduces a two-stage architecture that combines FerRCNN-based region processing with domain adaptation to focus on fog-relevant regions and suppress spurious areas. Key components include a domain discriminator, depth estimation block, reconstruction decoder, and pseudo-label generator to enforce cross-domain consistency and fog-invariant features, aiming to boost both accuracy and inference speed. Experimental results on a foggy road/vehicle dataset show an accuracy of about 85% and clear improvements over FerRCNN-based baselines, underscoring the method's potential for real-time autonomous driving in adverse weather.

Abstract

Object detection in reduced visibility has become a prominent research area. The existing techniques are not accurate enough in recognizing objects under such circumstances. This paper introduces a new foggy object detection method through a two-staged architecture of region identification from input images and detecting objects in such regions. The paper confirms notable improvements of the proposed method's accuracy and detection time over existing techniques.

Figures (5)

• Figure 1: Architecture of the proposed method.
• Figure 2: A sample foggy image from target domain.
• Figure 3: Object detection in image of Fig. \ref{['fig:1st_exp_orig']} by the proposed method.
• Figure 4: Another sample foggy image from target domain.
• Figure 5: Object detection in image of Fig. \ref{['fig:2nd_exp_orig']} by the proposed method.