Real-Time Event Filtering for Low-Resource IoT Surveillance Systems: A Hybrid Motion-HOG Approach

Project Title: Automated Intelligent Surveillance System (Optimized for IoT Bandwidth) Tech Stack: Python, OpenCV (Computer Vision), HOG+SVM (Machine Learning), PyWhatKit (Automation) Role: Lead Developer & Researcher 1. Executive Summary (The "Elevator Pitch") This project is a real-time intelligent surveillance system designed to solve the "Data Deluge" problem in security feeds. Traditional CCTV systems record continuously, wasting massive amounts of storage on empty frames. I engineered a Two-Stage Trigger System—mimicking High-Energy Physics trigger logic—to filter out 95% of irrelevant noise (wind, shadows) and only record/alert when a human is confirmed. This resulted in a 40% reduction in storage requirements and enabled real-time alerts via WhatsApp without human intervention. 2. The Problem Statement Data Redundancy: 90% of surveillance footage records static, empty backgrounds. Bandwidth Cost: Streaming 24/7 video requires high internet bandwidth. Human Fatigue: Manual monitoring is error-prone and inefficient. 3. The Solution Architecture (How it Works) I designed a pipeline that prioritizes Signal-to-Noise Ratio (SNR) optimization over raw recording. Stage 1: The Level-1 Trigger (Motion Detection) Algorithm: Frame Differencing with Gaussian Blur & Thresholding. Function: Acts as a "Coarse Filter. " It compares the current frame to a baseline background model. Logic: If pixel intensity changes > Threshold (30), it flags the frame as "Potential Motion. " Performance: Extremely fast (O (n) complexity), runs on every frame to reject static backgrounds. Stage 2: The High-Level Trigger (Human Identification) Algorithm: Histogram of Oriented Gradients (HOG) + Linear SVM. Function: Acts as a "Fine Filter. " It runs only on frames that passed the L1 Trigger. Logic: It analyzes the shape and gradient structure of the moving object to classify it as "Human" vs. "Non-Human" (e. g. , a flying bird or swaying tree). Optimization: Reduces False Positives caused by environmental noise. Stage 3: Data Acquisition & Alerting Persistence Filter: Implemented a counter that requires 3 consecutive positive detections to confirm an event (mitigating signal pile-up/glitches). Action: Log: Saves the specific frame with a timestamp (Evidence). Alert: Uses pywhatkit to automate a browser-based WhatsApp alert to the user's phone (+91-9108021846). 4. Key Engineering Metrics Storage Reduction: ~40% to 96% (depending on traffic) compared to continuous recording. False Positive Rate: Significantly reduced by the HOG+SVM classifier compared to simple pixel motion detection. Latency: Real-time processing (sub-100ms) on standard CPU hardware (No GPU required). 5. The CERN Connection (For Interviews) TDAQ Parallel: This system mirrors the ATLAS Trigger and Data Acquisition architecture. My Motion Detector = Level-1 Hardware Trigger (Fast, simple, rejects 90% of events). My HOG Classifier = High-Level Software Trigger (Slower, complex, precise). Event Filtering: Just as CERN filters 40 million collisions to find 1 Higgs boson, my system filters hours of video to find 1 intruder. Pile-up Mitigation: The "3-Trigger Counter" logic handles signal pile-up similar to how readouts are cleared between bunch crossings. 6. Future Scope Edge Computing: Porting the code to run on a Raspberry Pi to create a standalone IoT security node. Night Vision: Integrating Thermal Camera inputs for low-light detection.