Analog Gauge Digitization & VLM-Verified Threat Detection
Config-driven computer vision modules for industrial gauge digitization and security. Digitizes analog dial meters via Hough needle angle mapping and provides 3-stage threat detection with open-vocabulary YOLO and local Ollama VLM verification.
Internal R&D / Personal Engineering Project at Bondstein Technologies Ltd.
Module of Bondstein Vision Suite; presented as a separate case study.
Workflow at a glance
Acquire camera frames
Provide imagery to the selected task.
Select task workflow
Use gauge reading or threat-event detection.
Validate task signals
Use geometry or temporal/model review.
Review outputs
Inspect readings or flagged event evidence.
Project facts & reported results
Implementation fact
Implementation fact
Implementation fact
Project Overview
Problem Statement
Industrial facilities need repeatable analog-gauge readings and security alerts that separate real threats from tools and visually similar objects.
Approach & Methodology
Created an automated dial gauge reader using Hough transform line detection inside calibrated dial circles to map needle angles to engineering units. For security, developed a 3-stage threat detection pipeline: open-vocabulary YOLO -> temporal voting -> local Ollama VLM event review.
Outcome & Scope
Built configurable R&D modules for gauge digitization and multi-stage threat verification using local inference.
System Architecture
Industrial telemetry & multi-stage security verification engine
Key Components:
Finds meter needle angles and converts to calibrated engineering units
Identifies potential threat objects in video frames
Sliding-window voter filtering brief occlusion or detection flickers
Queries local vision-language model for secondary event assessment
Key Features & Capabilities
Analog gauge dial digitization via Hough needle angle detection & engineering unit mapping
Median-smoothed meter telemetry with alarm thresholding
3-stage weapon & threat detection: high-threshold detector -> temporal voter -> VLM confirmation
Local Ollama VLM integration for secondary event review to help reduce false-positive alerts
Production-grade vision_core runtime engine