Multimodal vision in and beyond the visible spectrum for security applications: wildland fires scenarios

近年来，人们对红外成像和多模式视觉的兴趣与日俱增；无论是来自研究界还是工业界。这种兴趣主要是由相机价格的下降和性能的提高推动的。曾经主要局限于军事的模式，现在正在扩展到商业和工业应用。这种扩展伴随着对新的高级信号和图像处理算法的需求，以解决各种新的新兴应用领域中的问题，使用可见光谱内外的多模式视觉。 近年来，在人脸识别、森林火灾研究、目标跟踪、无损检测等领域，发展了许多红外和多模式视觉算法。 这项工作旨在开发多模式视觉中的新算法，应用于安全应用，特别是非结构化和自然灾害场景中的多模式视觉系统的开发，如荒地和森林火灾。这项工作将在四个互补的轴线上进行，旨在建立一个更完整的火灾研究和消防多式联运系统： 1.在地面和空中飞行器中使用机器学习和多模式视觉系统对室外非结构化场景中的火灾进行检测和分割。 2.基于多模式协同视觉系统(地面和空中)的三维火灾特征提取。 3.地面和航空3D数据的配准。 4.火灾再燃的预测与检测。 这项工作将有助于增进我们在野火研究、多无人机(UAV)视觉系统以及应用于非结构化场景图像的深度学习领域的知识。开发的工具将进一步适应其他自然风险情景和搜救应用。科技成果将对社会经济和环境产生重要影响。通过这些发展，加拿大可以在应用于自然风险情景的技术领域发挥主导作用。最后，加强对高素质人才的培养。

Recent years have witnessed an increased interest in infrared imaging and multimodal vision; both from the research community and industry. This interest has been mainly driven by the decrease in camera prices and an increase in their performance. Modalities that were once mostly limited to the military, are now expanding into commercial and industrial applications. This expansion is accompanied by a demand for new advanced signal and image processing algorithms to solve problems in a variety of new emerging areas of application using multimodal vision in and beyond the visible spectrum. In recent years, many algorithms have been developed for infrared and multimodal vision in the areas of face recognition, forest fires study, target tracking, non-destructive testing, etc. This work aims to develop new algorithms in multimodal vision applied to security applications, in particular the developement of multimodal vision systems in unstructured and natural disasters scenarios such as wildland and forest fires. The work will be conducted in four complementary axes aiming to build a more complete fire study and firefighting multimodal system: 1. Detection and segmentation of fires in outdoor unstructured scenarios using machine learning and multimodal vision systems in ground based and aerial vehicles. 2. Extraction of three-dimensional fire characteristics using multimodal cooperative vision systems (ground and aerial). 3. Registration of ground and aerial 3D data. 4. Prediction and detection of fire reignition. This work will contribute to advancing our knowledge in the areas of wildland fires research, multi-UAV (Unmanned Aerial Vehicles) vision systems, and in deep learning applied to unstructured scenario images. The developed tools will be further adapted to other natural risk scenarios and search & rescue applications. The scientific and technological outcomes will have important socioeconomic and environmental impacts. Through these developments, Canada can play a leading role in the area of technologies applied to natural risk scenarios. Finally, this work will enhance the training of highly qualified personnel.