{"id":27641264,"url":"https://github.com/sayandeepmaity/luminator","last_synced_at":"2026-05-09T09:46:24.724Z","repository":{"id":289487985,"uuid":"969519382","full_name":"sayandeepmaity/luminator","owner":"sayandeepmaity","description":"Microphone Array-Based Direction of Arrival of Gunshot Detection .Gun violence remains a critical concern. Identifying the precise location of a gunshot—or getting as close as humanly possible—is crucial for saving lives and ensuring public safety.","archived":false,"fork":false,"pushed_at":"2025-04-23T14:53:52.000Z","size":112,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-04-23T15:22:02.977Z","etag":null,"topics":["classification","cnn","cst","deep-learning","fpga","gelu","gunshot-detection","neural-network","pandas","prediction","pytorch","regression","rnn","skicit-learn","tdoa","transformers","verilog-hdl"],"latest_commit_sha":null,"homepage":"","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/sayandeepmaity.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-04-20T10:38:01.000Z","updated_at":"2025-04-23T14:53:56.000Z","dependencies_parsed_at":"2025-04-23T15:22:05.334Z","dependency_job_id":"e03f0197-90d3-44c9-b559-c14fe5738e93","html_url":"https://github.com/sayandeepmaity/luminator","commit_stats":null,"previous_names":["sayandeepmaity/luminator"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sayandeepmaity%2Fluminator","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sayandeepmaity%2Fluminator/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sayandeepmaity%2Fluminator/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sayandeepmaity%2Fluminator/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/sayandeepmaity","download_url":"https://codeload.github.com/sayandeepmaity/luminator/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":250535069,"owners_count":21446503,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["classification","cnn","cst","deep-learning","fpga","gelu","gunshot-detection","neural-network","pandas","prediction","pytorch","regression","rnn","skicit-learn","tdoa","transformers","verilog-hdl"],"created_at":"2025-04-23T23:42:36.863Z","updated_at":"2026-05-09T09:46:19.673Z","avatar_url":"https://github.com/sayandeepmaity.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"# LUMINATOR\n**LIGHTWEIGHT ULTRA-ADVANCED MULTI-SENSOR INTELLIGENT NOISE-ABRIDGE THREAT OVERCOMING RADAR**\n\n---\n\n## Problem Statement Title\n**MICROPHONE ARRAY-BASED DIRECTION OF ARRIVAL OF GUNSHOT DETECTION**\n\nIt's important to know the **precise location** of a gunshot — or at least as close as humanly possible — because **gun violence** isn't going away anytime soon.  \nOur goal is to **harness any available technology** or assisting capability to ensure the **best possible outcome in worst-case scenarios.**\n\n---\n\n## Proposed Solution\n\nAn **advanced gunshot detection and classification system** using a **hybrid approach** of:\n- Microphone arrays\n- FPGA-based signal processing\n- Deep learning models\n\nBy leveraging **CNN-RNN neural architectures** and **Transformer layers**, we enable:\n- Real-time gunshot identification\n- 3D localization\n- Type classification\n\n---\n\n## How It Solves the Problem\n\n- Instant feedback with highly accurate 3D localization using TDOA\n- Multi-classification approach to detect and identify firearm types in real-world scenarios\n\n---\n\n## Role of Deep Learning\n\n- Distinguishes gunshots from other loud noises or distractions\n- Reduces false alarms\n- Ensures fast \u0026 accurate responses\n\n---\n\n### Key Features:\n- Immediate detection \u0026 3D localization\n- FPGA-based noise filtering and feature extraction\n- Deep learning-based classification\n- Real-time responsiveness\n\n---\n\n## Innovation and Uniqueness\n\n- Hybridization: Combining CNNs and RNNs for enhanced audio analysis\n- Transformer Integration: Improved attention to relevant features\n- Real-time Application: Live analysis of hazardous sound events\n\n---\n\n## Technologies Involved\n\n### Programming Languages\n- Python (ML Models)\n- Verilog (FPGA Processing)\n\n### Hardware\n- 4 or 6 omnidirectional microphones\n- FPGA for real-time signal processing\n- GPU/TPU for model training\n- LCD Display for output\n\n### Frameworks \u0026 Libraries\n- TensorFlow\n- PyTorch\n- Scikit-Learn\n- Pandas\n- TQDM\n\n### Machine Learning Models\n- CNN\n- RNN (LSTM)\n- Regression (TDOA)\n- CNN + Fully Connected\n- CST Transformer\n\n---\n\n## Methodology \u0026 Implementation\n\n1. Microphone Array captures the sound\n2. ADC converts it to digital signal\n3. FPGA applies bandpass filtering (up to 3kHz) \u0026 extracts features\n4. CNN-RNN Hybrid Model processes features\n5. CST Transformer Layer applies 3 attention types:\n    - Channel-MHSA: Spatial attention\n    - Spectral-MHSA: Frequency-based attention\n    - Temporal-MHSA: Time-based evolution attention\n6. GELU Activation enhances model performance\n7. Real-time output shown on display\n\n---\n\n## Technical Feasibility\n\n- Built on existing technologies like FPGAs and Neural Networks\n- Uses real-time signal processing and deep learning\n- Feasible and scalable with current hardware\n\n---\n\n## Why CNN + RNN Hybrid?\n\n### CNNs:\n- Extract local patterns and hierarchies (e.g., sound textures)\n- Reduce dimensionality while preserving important spatial info\n\n### RNNs:\n- Retain temporal memory\n- Model long sequences (like varying gunshot durations)\n- LSTM or GRU helps with temporal context in audio\n\n---\n\n## Model Performance Comparison\n\n| Model               | Gunshot Detected (%) | Muzzle Blast Detected (%) | Shockwave Detected (%) | TDOA Accuracy (%) | Gun Type Classification (%) |\n|--------------------|----------------------|----------------------------|-------------------------|-------------------|------------------------------|\n| CNN-Only           | 65                   | 83                         | 89                      | 72                | 80                           |\n| RNN-Only           | 83                   | 85                         | 94                      | 79                | 91                           |\n| DNN-Only           | 82                   | 90                         | 96                      | 77                | 93                           |\n| CNN+RNN            | 92                   | 93                         | 98                      | 82                | 95                           |\n| CNN+RNN+CST        | 96                   | 95                         | 97                      | 89                | 98                           |\n\n---\n\n## Challenges \u0026 Risks\n\n- Hardware Limitations: FPGAs have finite resources\n- Latency: Real-time needs demand high processing speed\n- Noise Interference: May affect accuracy\n- Data Availability: High-quality gunshot datasets are rare\n\n---\n\n## Optimizations \u0026 Techniques\n\n### FPGA Resource Optimization:\n- Designed efficient Verilog modules\n- Implemented parallel processing for real-time handling\n\n### Algorithm Improvements:\n- Developed real-time, low-latency algorithms for gunshot detection\n- Integrated high-pass/low-pass filters\n- Trained ML models with noise-augmented datasets\n\n---\n\n## References\n\n- [FPGA Filters](https://ashrafi.sdsu.edu/PDF/filters_FPGA.pdf)  \n- [Gunshot Detection Thesis - Auraria Library](https://digital.auraria.edu/files/pdf?fileid=e562b890-41f9-4170-a13b-27b07d1f1626)  \n- [TDOA Localization Field Guide](https://www.decodio.com/media/downloads/TDOA_Localization_From_Theory_to_the_Field_v1-0.pdf)  \n- [Multiple Impulse Acoustic Sources - MDPI](https://www.mdpi.com)  \n- [Gunshot Detection Using Accelerometers - PLOS ONE](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0199023)  \n- [Gun Identification using Transformer - Nature Scientific Reports](https://www.nature.com/articles/s41598-021-88594-3)  \n- [Raytheon's Boomerang Acoustic System](https://prd-sc102-cdn.rtx.com/-/media/rtx/our-company/who-we-are/bbn/boomerang/files/boomerang_data.pdf)  \n- [Object Tracking using TDOA - MATLAB](https://www.mathworks.com/help/dsp/ug/object-tracking-using-time-difference-of-arrival.html)  \n- [Gunshot-like Sounds Detection - MDPI](https://www.mdpi.com)  \n- [Postprint PDF - Google Drive](https://drive.google.com/file/d/1g3ceUoOC9cOvfQ/pdf)  \n- [FireBrick Project - Tufts University](https://sites.tufts.edu/eeseniordesignhandbook/files/2017/05/FireBrick_OKeefe_F1.pdf)  \n- [Indoor Gunshot Notification System - MDPI](https://www.mdpi.com)\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fsayandeepmaity%2Fluminator","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fsayandeepmaity%2Fluminator","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fsayandeepmaity%2Fluminator/lists"}