Industry engineering, internal R&D, and research in computer vision and machine learning
Showing 28 projects. Related system modules are listed separately.
Edge computer vision pipeline for reading ISO 6346 container numbers from reach stackers and side lifters using fisheye correction, YOLO-OBB detection, crop pairing, PaddleOCR, prefix correction, and check-digit validation.
Multiprocess RTSP analytics system that checks six leaf-buying SOP stages: bale opening, barcode scan, moisture inspection, layer-by-layer inspection, weighing step-back, and unwanted activities.
Multi-module safety platform covering vehicle overspeed estimation, PPE compliance, forklift-worker collision risk, restricted-area intrusion, unsafe behavior, and perimeter throwing detection.
Track-centric Bangla automatic number-plate recognition using YOLO-OBB detection, vehicle association, per-track crop queues, PaddleOCR, district and category validation, confidence voting, deduplication, and API delivery.
Staircase safety monitoring for handrail use and phone usage using YOLO person detection, MediaPipe Pose, dynamic handrail regions, wrist geometry, temporal stability, and evidence delivery.
Showroom analytics covering entry and exit, footfall, motorcycle movement and recognition, customer-product proximity, employee recognition, desk interaction, CRM capture, service delay, and shutter state.
Module of BRAC CIN
Identity-aware banking surveillance using InsightFace embeddings, FAISS similarity search, whitelist and blacklist categories, unknown-person enrollment, and a review workflow for registry quality.
Module of BRAC CIN
Configurable intrusion analytics supporting polygon occupancy and directional line crossing, with track-aware confirmation, cooldowns, evidence storage, and retryable API delivery.
Module of BRAC CIN
Pose-based monitoring of potentially unsafe counter interactions using customer and banker regions, counter-line geometry, role stabilization, temporal voting, evidence capture, and optional local vision-language review.
Module of BRAC CIN
Detects lying or fallen persons and policy-defined unusual activity using YOLO11 Pose, geometry scoring, track history, weighted temporal voting, cooldowns, and evidence deduplication.
Operator-facing visual QA application for product identification, label integrity checks, batch inspection, OCR watermark analysis, configurable markers, and genuine-versus-counterfeit review.
An on-premises video-intelligence R&D platform connecting camera monitoring, configurable ML tasks, spatial event rules, recordings, and operational health.
A LangGraph-based engineering prototype that organizes computer-vision briefs into plans and project scaffolds through specialist agents, human approval gates, and reusable templates.
Module of Bondstein Vision Suite
Industrial inspection R&D for blister-pack cavities, cap presence and seating, and liquid fill levels using lightweight geometry and image-processing methods with a queued inference runtime.
Module of Bondstein Vision Suite
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.
Module of Bondstein Vision Suite
Modular computer vision solutions for smart infrastructure: conveyor package directional counting with CentroidTracker, stall-level smart parking occupancy with polygon hysteresis, pyzbar catalog verification, and open-vocabulary fire/smoke detection.
Multi-modal public safety analytics system combining YOLO multi-class detection, DeepSORT tracking, depth estimation, and heatmap smoothing for crowd monitoring, threat detection, and weapon-aware escalation.
U-Net segmentation R&D for inspecting juice-flow imagery, with temporal confirmation and a planned Jetson Orin Nano deployment path.
Visual inspection R&D for distinguishing genuine and counterfeit packaging using learned features, OCR, and classical image-quality checks.
Internal HR service-desk implementation combining cited policy answers, role-based self-service, approval workflows, SLA tracking, email operations, and audit logs.
CNN–Vision Transformer research for binary classification of Bengali voice samples, published in Journal of Voice.
Comparative research on VGG16, InceptionV3, and DenseNet121 for voice-spectrogram classification using augmentation and transfer learning.
Hybrid DenseNet–ViT–GRU architecture with Explainable AI (Grad-CAM, LIME, SHAP) for diagnosing psychological stability from voice spectrograms. Combines local, global, and sequential feature learning with transparent, interpretable predictions.
Self-supervised pretraining with masked spectrogram reconstruction followed by supervised ViT-style classification for robust mental stability detection from noisy, unprocessed real-world voice data.
Research on benign/malignant skin-lesion classification using HAM10000 and a custom DCNN, with saliency and attention visualizations. Presented as an arXiv preprint.
Academic proof of concept combining permissioned health-record workflows, smart contracts, and experimental AI-assisted analysis.
Machine learning-based caloric estimation system for traditional Bangladeshi dishes to promote nutritional awareness.
Large-scale machine learning models to predict physical activity levels from health parameters for lifestyle intervention.