Future developments involve cloud

The gap

Evidence profile

Sourced from the future work and recommendations of the source papers, classified as general, spanning 4 journals.

Supporting evidence — 4 representative gaps

• IoT-Enabled Speed and Accident Detection Platform Using Deep Learning and Multi-Object Tracking (2026) · Open Access Journal on Engineering Applications · doi

  Several directions for future enhancement are identified: 1. Accident Detection: Incorporating LSTM or transformer-based temporal sequence models to detect abnor- mal vehicle behavior patterns—including sudden stops, abrupt lane changes, and collision trajectories—to enable proactive accident detection and automated emergency dispatch. 2. Multi-Modal IoT Sensing: Integrating IoT sensors (accelerometers, GPS modules, vibration sensors) with the video analytics pipeline for multi-modal accident confirmation, improving both detection reliability and geo-location precision. 58 Open Access Journal on Engineering Applications DOI:10.64886/oajea.0102.006 3. Cloud-Scale Deployment: Extending the MQTT IoT layer to support multi-camera deployments connected to centralized cloud dashboards, enabling city-wide real-time traffic monitoring, congestion prediction, and automated traffic signal control. 4. Nighttime and Adverse Weather Robustness: Incorporating infrared or thermal camera inputs, along with domain adaptation techniques, to maintain detection and tracking accuracy under low illumination, rain, and fog conditions. 5. Edge Deployment: Porting the YOLOv8 inference to edge AI hardware (NVIDIA Jetson, Google Coral) to enable fully self-contained, low-latency, and low-power deployment at roadside locations without reliance on central GPU servers. 6. License Plate Recognition: Integrating Optical Character Recognition (OCR) for license plate reading to enable automated violation record creation and linkage to traffic authority databases.

  generalfuture workevidence 5/5
  Keywords: detection accident enable automated multi deployment traffic incorporating modal integrating sensors cloud camera edge license
• Multimodal Computer Vision for Rapid Disaster Damage Analysis and Victim Detection using Deep Learning (2026) · Zenodo (CERN European Organization for Nuclear Research) · doi

  The current implementation establishes a functional baseline for hybrid disaster scene analysis. Several directions are identified for future development. Domain-Specific Model Fine-Tuning: Training YOLOv8 or YOLO11 variants on annotated disaster- scene datasets including fire, smoke, rubble, and collapsed structures would replace the HSV and edge density modules with more precise neural alternatives. Public datasets such as AIDER (Aerial Image Dataset for Emergency Response) and the Foggia fire detection dataset could serve as training foundations. Quantized Neural Networks: Applying post- training quantization (INT8) to the YOLOv8n-seg model using tools such as ONNX Runtime or TensorRT (if future deployments include GPU hardware) would further reduce inference latency, potentially enabling 30 FPS victim tracking on current CPU hardware. Temporal Object Tracking: a tracker such as SORT or lightweight multi-object identity ByteTrack would enable consistent victim tracking across frames, providing trajectory information useful for rescue coordination.

  generalfuture workevidence 5/5
  Keywords: training tracking current disaster scene future model yolov datasets fire neural dataset hardware victim object
• GNSS and multi-modal data integration methods for landslide displacement monitoring: a review (2026) · Geohazards & Remediation · doi

  Future developments involve cloud–edge co-process- ing and low-power wide-area networks (LPWANs) for IoT. To scale up monitoring in mountainous regions, it is recommended to adopt edge intelligence for low-latency preprocessing (time sync, denoising, outlier removal, and  baseline-shift mitigation) and on-device anomaly screening, while cloud platforms handle model train- ing, multi-sensor fusion, and data archiving. In addition, constructing digital twins of landslides based on physics- informed and data-assimilative models can bridge the gap between observational and computational agents, providing an intelligent, interpretable module for mecha- nistic explanation and operational decision-making.

  generalrecommendationsevidence 5/5
  Keywords: cloud edge future developments involve process power wide area networks lpwans scale monitoring mountainous regions
• AI-Based Accident Severity Ranking and Emergency Response Automation System (2026) · International Journal of Science and Research (IJSR) · doi

  This work presents an end-to-end accident severity ranking and emergency response automation framework integrating image preprocessing, contextual scene understanding, license-plate extraction, deep-learning severity classification, severity-based decision making, email notification, and GUI- based result presentation. The Phase 2 MobileNetV3 implementation reports 93.35% accuracy, 94.26% precision, 93.35% recall, and 93.80% F1-score, improving substantially over the Phase 1 MobileNetV2 baseline. The integration of Gemini-3.6-Flash adds human-readable accident context such as vehicle damage, vehicle count, persons, fire, smoke, and collision characteristics, which complements the numerical severity prediction. and availability, privacy, Future work will extend the current image-based system to live dashcam or CCTV video, investigate an offline VLM for improved cloud dependency, incorporate vehicle-specific and NCAP-related safety information into severity assessment, and add SMS and WhatsApp multi-channel communication. Further evaluation should also report class- wise confusion-matrix values, plate-recognition accuracy, structured VLM field correctness, notification-delivery success rate, and end-to-end latency on a fixed test protocol. alongside reduced email for

  generalfuture workevidence 5/5
  Keywords: severity based vehicle accident image plate email notification phase mobilenetv accuracy presents ranking emergency response

Questions about this gap