SureView Solutions LLC runs real-time detection on GDOT camera streams. Vehicles are tracked with persistent IDs to reduce flicker/ID-switching, then scored for collision-like interactions. When thresholds are crossed, we flag the event, capture buffered video (pre/post), and ship it to cloud storage with full metadata.
Configurable IOU and cosine-similarity thresholds differentiate fender-benders from major crashes. Alerts flow to email, SMS, or webhook receivers (your CAD/dispatch or custom API).
Our pipeline pairs YOLOv8 detections with a robust tracker to maintain identity through occlusions. We persist events as MP4 clips, save JSON metadata (IDs, vectors, IOU snapshots), and optionally redact plates on export. Clips are securely stored in the cloud.
Deploy cloud-native on GCP with VPU-accelerated inference, or at the edge. 24/7 operation, health checks, and automatic restarts keep feeds hot.
1) YOLOv8 detects vehicles; tracker assigns stable IDs.
2) Heuristics label fender vs crash; pre/post buffers roll.
Differentiate fender-benders from major crashes with tunable IOU and trajectory heuristics.
Stable track IDs reduce ID-switching across occlusions and camera jitter to cut false alarms.
Pre/Post buffering wraps the full incident window. H.264 MP4 with timestamped filenames.
OCR for plate text/state when permitted; automatic redaction on exports to protect privacy.
Email/SMS/webhook notifications. Integrate with CAD/dispatch or your internal APIs.
Search events by time/location, stream clips, and pull metadata via a clean REST API.
