Internship Experience

Project: Vehicle Tracking System With Computer Vision and Machine Learning

Role: AI & ML Developer Intern

Company: Airpix Geoanalytics, based in Mumbai, India, specializes in AI-powered intelligent video surveillance and analytics solutions, offering services that are deployable on the edge, cloud, or hybrid platforms.

Tasks:

• Develop an application for vehicle counting, classification, and tracking based on Indian vehicle categories.
• Implement traffic violation detection, including stop line crossing and traffic sign/signal infractions.
• Integrate with the company’s analytics website for data retrieval and requests.
• Deploy on NVIDIA Jetson edge devices (Xavier, Nano, AGX).

My Contributions:

• Implemented video analytics using computer vision and machine learning models for real-time vehicle identification, classification, and tracking from IP camera feeds.
• Utilized SSD frameworks for object detection and DeepSORT for vehicle tracking, incorporating asynchronous programming (asyncio) for concurrency.
• Developed a multiprocessing and multithreaded system to handle complex operations simultaneously, including a software watchdog for system reliability.
• Optimized and deployed the solution on Edge device - NVIDIA Jetson Xavier using TensorRT for efficient inference on GPUs.
• Integrated the trained model with the backend and frontend systems, enabling real-time updates and adjustments through the dashboard.
• Successfully deployed the system at toll plazas, reducing vehicle waiting times by 42%.

Frameworks and Technologies

Python, C++, TensorRT, Cuda, OpenCV, Asyncio, SSD, Yolo, DeepSORT, Node.js and Systemd.

Concept

Object Detection/Classification/Tracking, Synchronous/Asynchronous programming, Multiprocessing & Multithreading, Software & Hardware Watchdogs.