Color–depth multi-task learning for object detection in haze

Haze environments pose serious challenges for object detection, making existing methods difficult to generate satisfied results. However, there is no escape from haze environments in real-world applications, especially in water and bad weather. Hence, it is necessary to enable object detection metho...

Full description

Saved in:
Bibliographic Details
Published in:Neural computing & applications 2020-06, Vol.32 (11), p.6591-6599
Main Authors: Chen, Zhe, Wang, Xin, Fan, Tanghuai, Xu, Lizhong
Format: Article
Language:English
Subjects:
Artificial Intelligence
Color
Computational Biology/Bioinformatics
Computational Science and Engineering
Computer Science
Data Mining and Knowledge Discovery
Haze
Image Processing and Computer Vision
Learning
Meteorological satellites
Object recognition
Probability and Statistics in Computer Science
S.I. : Multi-Source Data Understanding (MSDU)
Telematics
Weather
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Haze environments pose serious challenges for object detection, making existing methods difficult to generate satisfied results. However, there is no escape from haze environments in real-world applications, especially in water and bad weather. Hence, it is necessary to enable object detection methods to conquer the difficulties caused by the haze effect. In spite of the diversity between various conditions, haze environments share a common characteristic that the haze concentration is changed with the scene depth. Hence, this haze concentration feature can be used as a representation of the scene depth. This provides us a novel cue available for object detection in haze that the object-background depth contrast can be identified. In this paper, we propose a multi-task learning-based object detection method by jointly using the color and depth features. A pair of background models is built separately with the color and depth features, forming two streams of our multi-task learning framework. The final object detection results are generated by fusing the results given by color and depth features. In contrast to existing object detection methods, the novelty of our method lies in the combination of the color and depth features under a unified multi-task learning mechanism, which is experimentally demonstrated to be robust against challenging haze environments.
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-018-3732-6