Gaseous Object Detection

TL;DR

This paper defines Gaseous Object Detection (GOD) as a distinct vision task characterized by saliency-deficient, arbitrarily shaped, and boundary-sparse gas plumes. It introduces the GOD-Video dataset (600 videos, 141,017 frames) and a physics-inspired approach, the voxel shift field (VSF), integrated into Faster RCNN (VSF RCNN) to capture 3D spatio-temporal cues via Gaussian dispersion-based offsets. VSF RCNN achieves substantial improvements over baselines (e.g., $AP_{50}$ from 36.15% to 51.08% and overall $AP$ to 20.43%), validating the effectiveness of joint data- and feature-level temporal modeling for gaseous objects. The work offers a robust dataset, comprehensive benchmarks, and a flexible, generalizable baseline that can be extended to other detectors and gas-imaging applications, driving future research in gas-centric perception and multi-dimensional detection."

Abstract

Object detection, a fundamental and challenging problem in computer vision, has experienced rapid development due to the effectiveness of deep learning. The current objects to be detected are mostly rigid solid substances with apparent and distinct visual characteristics. In this paper, we endeavor on a scarcely explored task named Gaseous Object Detection (GOD), which is undertaken to explore whether the object detection techniques can be extended from solid substances to gaseous substances. Nevertheless, the gas exhibits significantly different visual characteristics: 1) saliency deficiency, 2) arbitrary and ever-changing shapes, 3) lack of distinct boundaries. To facilitate the study on this challenging task, we construct a GOD-Video dataset comprising 600 videos (141,017 frames) that cover various attributes with multiple types of gases. A comprehensive benchmark is established based on this dataset, allowing for a rigorous evaluation of frame-level and video-level detectors. Deduced from the Gaussian dispersion model, the physics-inspired Voxel Shift Field (VSF) is designed to model geometric irregularities and ever-changing shapes in potential 3D space. By integrating VSF into Faster RCNN, the VSF RCNN serves as a simple but strong baseline for gaseous object detection. Our work aims to attract further research into this valuable albeit challenging area.

Figures (17)

• Figure 1: We compare the conventional object and gaseous object using traditional feature descriptors in three aspects: saliency, shape and boundary. Higher scores of the objectness measures indicate that the box area is more likely to contain an object, while the gaseous object shows greater similarity to the background rather than the foreground.
• Figure 2: The spectral transmittance curves of representative gases in the mid-infrared band. Despite differences in gas types and spectral ranges, based on the Lambert-Beer's law, they exhibit similar visual characteristics in gas imaging cameras.
• Figure 3: We show the representative samples of different attributes in the GOD-Video dataset. The green rectangles represent the annotated boxes.
• Figure 4: GOD-Video dataset details. (a) Mutual dependencies among attributes. (b) Scene taxonomy of the GOD-Video dataset (SO: Small Object, MO: Middle Object, BO: Big Object). (c) Comparisons with previous gas leak detection datasets in the frame-level. The quantity of samples in the GOD-Video dataset surpasses previous datasets significantly.
• Figure 5: The visualization of objectness measures (CC, MS, SS, ED) with corresponding scores, and the visualization of HOG for the train and gas samples.