TL;DR
This paper systematically reviews the use of computer vision and IoT-based sensors for flood monitoring and mapping, highlighting their combined potential for early warning systems. It identifies key challenges and proposes future research directions, particularly for under-explored areas like coastal lagoons.
What Problem Does This Paper Address?
Floods are a devastating natural hazard causing significant loss of life and economic damage globally. Effective monitoring and early warning systems are crucial for mitigating these impacts. While traditional methods exist, the increasing availability of data from IoT sensors and advancements in computer vision offer new opportunities for more robust and cost-effective flood management. However, the literature on how to best integrate these technologies, especially for complex environments like coastal lagoons, is fragmented. This paper aims to consolidate and critically appraise existing research to identify current trends, gaps, and future research avenues.
Methods
This paper presents a systematic literature review following PRISMA guidelines. The authors searched three major databases (Scopus, IEEE Xplore, Science Direct) using keywords related to remote sensing, IoT, flood monitoring, and computer vision. They screened over 13,000 records, ultimately selecting 91 relevant articles published between 2009 and 2019. The review categorizes and analyzes the methodologies employed, focusing on computer vision techniques (e.g., image processing, deep learning like CNNs) and IoT-based sensor approaches (e.g., pressure transducers, ultrasonic sensors, WSNs) for flood monitoring, mapping, and early warning. The paper also discusses data fusion techniques that combine data from multiple sources.
Datasets
The paper discusses various datasets used in the reviewed studies. Some studies utilized existing datasets, such as those for image segmentation of water bodies or geo-referenced labeled images from satellites (e.g., Sentinel-2). Others relied on crowdsourced social media images and textual data for training machine learning models. Some research also involved collecting data from ground-based cameras, UAVs, and IoT sensors in real-world scenarios. The paper mentions the creation of a Tiramisu image segmentation dataset and a EuroSAT dataset for land use and land cover classification, which can be relevant for flood-related applications.
Key Insights and Ideas
- Synergy is Key: The combination of computer vision (offering broad coverage and visual context) and IoT sensors (providing accurate, point-based measurements) is crucial for robust flood monitoring. Neither technology alone is sufficient.
- Deep Learning Dominance: Convolutional Neural Networks (CNNs) are increasingly prevalent and show superior performance in tasks like image segmentation and flood detection compared to traditional image processing techniques.
- UAVs as Game Changers: Unmanned Aerial Vehicles (UAVs) offer a flexible, cost-effective, and rapid way to collect high-resolution data for flood mapping and monitoring, especially in hard-to-access areas.
- Data Fusion for Accuracy: Combining data from diverse sources (e.g., satellite imagery, sensor networks, DEMs, GIS) significantly improves the accuracy and reliability of flood prediction and mapping models.
- Coastal Lagoons: An Underexplored Frontier: There's a significant research gap in applying computer vision and IoT sensors to monitor and manage coastal lagoons, which are vulnerable to flooding due to their unique hydrological characteristics.
Key Results
The review highlights that computer vision techniques, particularly CNNs, achieve high accuracy (e.g., 98.57% for ResNet-50 in RGB band combination) in tasks like water body segmentation. IoT-based sensor approaches are noted for their accuracy in water level estimation, though they are limited to point-based readings. The paper analyzes various studies, categorizing their accuracy and generalization capabilities, with many achieving "good" to "state-of-the-art" performance in specific applications. For instance, some UAV-based flood detection methods report accuracies up to 99.12%.
Limitations and Future Work
The authors identify several limitations and propose future research directions:
- Physical Measurement Dependency: Many computer vision techniques still rely on physical measuring scales (like staff gauges), which can be removed by integrating advanced CNNs with water level sensors.
- Environmental Variations: Traditional image processing techniques struggle in dynamic environments; deep learning models are recommended to overcome this.
- Limited Site Coverage: Future work should focus on improving the generalizability of models to various sites and conditions.
- Data Scarcity: The lack of open-source datasets for training computer vision algorithms is a significant challenge. Future research should focus on creating and sharing such datasets.
- Coastal Lagoon Focus: A major recommendation is to direct future research towards developing and applying computer vision and IoT sensor technologies for monitoring and managing coastal lagoons, including berm height estimation and automated control of lagoon entrances.
- Data Fusion for Validation: Further exploration of data fusion techniques is encouraged to reduce false positives in water level readings and enhance overall monitoring accuracy.
Summary updated