Edge-Based Standing-Water Detection via FSM-Guided Tiering and Multi-Model Consensus

Oliver Aleksander Larsen; Mahyar T. Moghaddam

cs.CVMar 31, 2026

Edge-Based Standing-Water Detection via FSM-Guided Tiering and Multi-Model Consensus

Authors: Oliver Aleksander Larsen, Mahyar T. Moghaddam

Organizations: SDU – University of Southern Denmark, MMMI, Odense, Denmark

Abstract

Standing water in agricultural fields threatens vehicle mobility and crop health. This paper presents a deployed edge architecture for standing-water detection using Raspberry-Pi-class devices with optional Jetson acceleration. Camera input and environmental sensors (humidity, pressure, temperature) are combined in a finite-state machine (FSM) that acts as the architectural decision engine. The FSM-guided control plane selects between local and offloaded inference tiers, trading accuracy, latency, and energy under intermittent connectivity and motion-dependent compute budgets. A multi-model YOLO ensemble provides image scores, while diurnal-baseline sensor fusion adjusts caution using environmental anomalies. All decisions are logged per frame, enabling bit-identical hardware-in-the-loop replays. Across ten configurations and sensor variants on identical field sequences with frame-level ground truth, we show that the combination of adaptive tiering, multi-model consensus, and diurnal sensor fusion improves flood-detection performance over static local baselines, uses less energy than a naive always-heavy offload policy, and maintains bounded tail latency in a real agricultural setting.

Figures & tables

Fig. 1 : High-level logical view of the three-node flood-detection architecture

Fig. 3 : High Level Motion-aware finite-state machine

Ensemble Slot	Dataset	Images	mAP50
Model 1	UB Water Detection	2,902	92.2%
Model 2	SegWater	1,963	62.8%
Model 3	Ponding-v2	4,768	89.9%

TABLE I : Three-model YOLOv8 ensemble used for consensus aggregation. Datasets: UB Water Detection [ 12 ] , SegWater [ 13 ] , and Ponding-v2 [ 14 ] .

Condition	Threshold	Boost
Humidity rise + cooling	$\Delta$ RH $>$ 15% and $\Delta T$ $<$ –1.5°C	+0.08
Temperature drop	$\Delta T$ $<$ –2.5°C	+0.04
Pressure fall	$\Delta P$ $<$ –5 hPa	+0.02
Hot & very dry	$\Delta$ RH $<$ –20% and $\Delta T$ $>$ +3.0°C	–0.08

TABLE II : Sensor boost rules motivated by flash-flood meteorological precursors (Doswell et al. [ 15 ] ) and suppressed runoff under dry antecedent conditions (Wasko and Nathan [ 16 ] ).

Sequence ID	Motion Label	Vehicle Behaviour
slow_creeping	slow	Passing flooded section
fast_passing	fast	Toward/away from hazard
stopped_water	stopped	Stationary at water
stopped_water_2	slow	Around constant flooding
slow_no_water	slow	No water, fields only

TABLE III : Test sequences and motion labels.

ID	Models/Tiers	FSM	Sensor Fusion	Offload Policy	Jetson	Purpose
1	1 $\times$ small	Off	Off	No	No	Static single-model upper bound
1b	1 $\times$ nano	Off	Off	No	No	Realistic static local baseline
2	3 $\times$ n/s/m/l	On	Off	Adaptive	Yes	Vision-only FSM with consensus
2b	1 $\times$ n/s/m/l	On	Off	Adaptive	Yes	Consensus ablation in vision FSM
3	3 $\times$ nano	On	On	No	No	Full system without Jetson
3b	1 $\times$ nano	On	On	No	No	Consensus ablation local-only

TABLE IV : Ablation configurations.

Fig. 4 : Adaptive behaviour across hazard severity: grouped bars for energy (left axis) and heavy-tier usage (right axis) by regime for baselines and production models.

ID	Configuration	Total Energy (J)	Macro F1 (2-class)	Balanced Accuracy	p99 Lat. (ms)	Temporal Coverage
1	static_small	171.7	0.755	0.733	4913.4	0.412
1b	static_nano	161.3	0.627	0.629	2552.3	0.470
2	vision_fsm_multi	262.0	0.785	0.789	2587.6	0.819
2b	vision_fsm_single	265.5	0.783	0.785	1137.2	0.935
3	full_local_multi	167.6	0.740	0.726	2638.0	0.462
3b	full_local_single	167.3	0.638	0.639	1209.8	0.575

TABLE V : Overall performance across the full test set.

Variant	Frames	Temp. Cov.	Flood R.	Watch R.
Neutral	21	0.667	0.917	0.286
Real-wet	29	0.917	0.923	0.833
Anti-flood	20	0.667	1.000	0.167

TABLE VI : Effect of sensor fusion.

Computation Offloading

See all 15 papers

Edge-Efficient Two-Stream Multimodal Architecture for Non-Intrusive Bathroom Fall Detection
Mar 17, 2026Haitian Wang, Yiren Wang, Xinyu Wang +2Fall Detection Multimodal Sensor Fusion
The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia
A Comprehensive Evaluation of Deep Learning Object Detection Models on Heterogeneous Edge Devices
Sep 25, 2024Daghash K. Alqahtani, Muhammad Aamir Cheema, Maria A. Rodriguez +1Energy-Efficient ML Edge Inference
DisNet Lab, School of Computing and Information Systems, The University of Melbourne, Melbourne, VIC, Australia · Monash University, Melbourne, VIC, Australia
Cross-Architecture Foundation-Model Distillation for Edge Flood Segmentation
Sep 17, 2026Fabian Schmalstieg, Karsten Mueller, Wojciech SamekRemote Sensing Image Segmentation Geospatial Foundation Models
Department of Artificial Intelligence, Fraunhofer Heinrich Hertz Institute, 10587 Berlin, Germany

Mar 17, 2026J/K move · Enter open · S save

Edge-Efficient Two-Stream Multimodal Architecture for Non-Intrusive Bathroom Fall Detection

Haitian Wang, Yiren Wang, Xinyu Wang, Sheldon Fung+1

The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia

Fall Detection Multimodal Sensor Fusion Human Activity Recognition Edge Inference mmWave Radar Sensing

Open paper AskPDF arXiv

Falls in wet bathroom environments are a major safety risk for seniors living alone. Recent work has shown that mmWave-only, vibration-only, and existing multimodal schemes, such as vibration-triggered radar activation, early feature concatenation, and decision-level score fusion, can support privacy-preserving, non-intrusive fall detection. However, these designs still treat motion and impact as loosely coupled streams, depending on coarse temporal alignment and amplitude thresholds, and do not explicitly encode the causal link between radar-observed collapse and floor impact or address timing drift, object drop confounders, and latency and energy constraints on low-power edge devices. To this end, we propose a two-stream architecture that encodes radar signals with a Motion--Mamba branch for long-range motion patterns and processes floor vibration with an Impact--Griffin branch that emphasizes impact transients and cross-axis coupling. Cross-conditioned fusion uses low-rank bilinear interaction and a Switch--MoE head to align motion and impact tokens and suppress object-drop confounders. The model keeps inference cost suitable for real-time execution on a Raspberry Pi 4B gateway. We construct a bathroom fall detection benchmark dataset with frame-level annotations, comprising more than 3~h of synchronized mmWave radar and triaxial vibration recordings across eight scenarios under running water, together with subject-independent training, validation, and test splits. On the test split, our model attains 96.1% accuracy, 94.8% precision, 88.0% recall, a 91.1% macro F1 score, and an AUC of 0.968. Compared with the strongest baseline, it improves accuracy by 2.0 percentage points and fall recall by 1.3 percentage points, while reducing latency from 35.9 ms to 15.8 ms and lowering energy per 2.56 s window from 14200 mJ to 10750 mJ on the Raspberry Pi 4B gateway.

Abstract

Computation Offloading

Explore similar work

Edge-Efficient Two-Stream Multimodal Architecture for Non-Intrusive Bathroom Fall Detection

A Comprehensive Evaluation of Deep Learning Object Detection Models on Heterogeneous Edge Devices

Cross-Architecture Foundation-Model Distillation for Edge Flood Segmentation