Archana Adhi

1. Sign Language Detection using Custom CNN Model

• Developed a custom CNN for real-time sign language recognition with 95% accuracy using TensorFlow, Keras, and OpenCV for video stream processing
• Ensured robustness and high performance through extensive validation and testing across various environments

2. Targeted Marketing Model Development for SMARTMARKET

• Engineered a machine-learning model to optimize campaign delivery, identifying high-probability responders using classification algorithms
• Reduced email marketing costs by targeting 25% of the subscriber base with high engagement likelihood

3. Aircraft Tracking using PTZ Cameras

• Developed a real-time aircraft tracking system using OpenCV and YOLO, deployed on NVIDIA Jetson for efficient video processing
• Integrated PTZ camera control for automatic aircraft tracking and created a web interface for live monitoring and data analysis, enhancing airport surveillance

4. 3D Cell Tower Modelling and AI Integration

• Developed an application using Structure-from-Motion algorithms for precise 3D modelling of cell towers
• Integrated advanced AI models on NVIDIA's Jetson platform for real-time processing and structural analysis

5. Geo-referencing of Mobile LiDAR Data

• Led the Geo-referencing initiative, aligning point cloud data within real-world spatial reference frameworks
• Authored Python scripts to fuse INS data with LiDAR data, producing highly accurate Geo-referenced point clouds

6. Advanced Point Cloud Manipulation

• Utilized Point Cloud Library, Open 3D, and PDAL for state-of-the-art point cloud processing
• Enhanced data quality through techniques like clustering, segmentation, filtering, and noise removal

7. Mobile Survey System Desktop Application Development

• Engineered a desktop application using pyQt for processing log files from mobile survey systems
• Enabled generation of trajectory files in custom formats and conversion of panoramic images to JPEG

8. Ground Control Station for Gimbal Cameras

• Developed a Ground Control Station using C++ and Qt, facilitating advanced surveillance capabilities with day and IR cameras

9. Deep Learning Object Detection Models

• Developed and deployed custom deep learning models tailored to client-specific requirements, leveraging dedicated GPUs and edge devices for optimal performance
• Created models capable of detecting individuals using PTZ camera feed at distances up to 800 meters with IR and up to 5000 meters with RGB, achieving 29 FPS and a mean Average Precision (mAP) of 0.90.

• Built quadcopters using the open-source pixhawk hardware and PX4 flight stack
• Worked on implementing robot localization in virtual environment (RViz)