Computer Vision C++

DENESH NALLUR NARASIMMAN

*******@***.*** +1-202-***-**** GitHub: deneshn LinkedIn: Denesh Narasimman Santa Clara, California, USA EDUCATION

University of Maryland, College Park

Master’s in Engineering, Robotics

Geethanjali College of Engineering and Technology, India Bachelor of Technology, Mechanical Engineering

SKILLS

Programming Languages: Python, C++

Technologies, Libraries and Tools: ROS, Gazebo, OpenCV, PCL, TensorFlow, PyTorch, Git, SolidWorks, AutoCAD, MATLAB, Arduino IDE, ANSYS, Creo, Linux, AWS, Linux, SLAM, sensor integration, RviZ, Point Cloud data simulation, Golang Machine Learning and Deep Learning Architectures: CNN, RNN, VGG16, ResNet, DenseNet, Transformers, Sklearn, LLM, NLP WORK EXPERIENCE

Swift Global Systems Robotics Engineer

Maryland, USA Aug 2023 - Present

• Developed and refined advanced computer vision algorithms using Python and C++ to enhance automated visual recognition systems. Achieved a 40% improvement in object detection accuracy by integrating and fine-tuning deep learning models in TensorFlow and PyTorch.

• Enhanced the performance of vision system by optimizing AI-driven image processing algorithms for real-time applications using efficient coding practices and reduced computational latency by 30%.

• Led the integration of computer vision algorithms into existing robotic systems, improving overall system efficiency by 20%. Employed distributed computing techniques to manage large-scale data processing, ensuring seamless functionality across complex robotic operations.

• Established rigorous testing protocols and benchmarks specifically for computer vision models to evaluate their performance under various operational conditions. Increased testing throughput and accuracy by 25%, directly contributing to faster iterations and enhancements.

• Engaged in research focused on novel applications of computer vision in robotics. Documented and shared innovative findings through internal reports and presentations, fostering knowledge growth within the engineering team. University of Maryland, College Park Research Assistant Maryland, USA Jan 2023 – May 2023

• Developed and optimized a Q-learning algorithm to enhance camera-based navigation systems for robotics, achieving a 40% improvement in obstacle avoidance accuracy.

• Conducted simulations in a cluttered environment using a Turtlebot, employing Python and ROS to navigate and avoid obstacles. Leveraged reinforcement learning techniques for real-time data processing and analysis.

• Achieved a 35% improvement in the autonomous learning capabilities of Turtlebots over training periods, significantly enhancing their proficiency in obstacle avoidance tasks.

Indian Institute of Technology, Hyderabad Deep Learning Research Assistant Hyderabad, India Jan 2021 – Aug 2021

• Facilitated 8-month project on small remotely piloted drone using PX4-Autopilot.

• Achieved seamless integration of hardware, software, and sensors for synchronized system.

• Visualized the environment using RViZ and LiDAR point cloud data for testing and development.

• Simulated drone performance in Gazebo, resulting in 15% longer flight time and improved stability.

• Coordinated with 3 teams, reducing integration time and improving system performance. PROJECT EXPERIENCE

Robot Perception :: Pedestrian Detection in Rainy Conditions using YOLO v4

• Implemented YOLO v4, a real-time object detection model utilizing convolutional neural networks (CNN), integrating image processing techniques for accurate pedestrian detection.

• Employed data augmentation techniques such as image rotation, scaling, and brightness adjustment to further enhance the diversity and generalization capabilities of the pedestrian detection model.

• Tailored dataset to include diverse weather conditions, enhancing robustness of pedestrian detection algorithm. Robot Motion Planning :: Path planning for multi-Agent Kilobots

• Orchestrated a multifaceted project, overseeing kilobot commissioning, literature review, and the precise assembly of 4 Arduino overhead controllers. Executed pattern formation algorithms with precision.

• Diligently investigated and resolved issues stemming from outdated kilobot libraries and limited literature, dedicating significant time to ensure comprehensive problem-solving and knowledge acquisition.

• Demonstrated proficiency in kilobot hardware, conducting 30 successful experiments and optimizing 4000 lines of code, resulting in a remarkable 45% boost in workflow efficiency. Deep Learning and Artificial Intelligence :: Action Recognition using 3D CNN based Spatio-temporal recognition

• Improved video action recognition by implementing a 3D CNN with Bidirectional Encoder Representations Transformer

(BERT), substituting the Temporal Average Pooling Layer.

• Resulted in a remarkable 10% accuracy and efficiency enhancement, accompanied by significant GFlops computational efficiency gains over the base model.

TRAINING AND CERTIFICATION

TensorFlow Certification Program :: TensorFlow Developer Certificate Issued September 2023 – Expires September 2026

Contact this candidate