Detail Information
Company Name Ericsson India Private Limited
Role/Position Machine Learning Intern
Duration May 2024 - June 2024
Location Noida, India
Certificate ‣
Tech Stack Python, OpenCV, NumPy, YOLO, Deep Learning, Transfer Learning, pyttsx3, JSON, Arduino, Raspberry Pi, Camera Modules, LiDAR Sensors, Ultrasonic Sensors, Microphones, Speaker Modules
Company Website https://www.ericsson.com/en
Final Presentation ‣

Drishti - 360° Vision Headband

Problem Statement

Modern-day safety challenges, especially in environments with low visibility (fog, crowd, etc.), pose a risk to individuals navigating such scenarios. Visually impaired individuals, delivery personnel, and those in rescue operations need real-time environmental awareness to make informed decisions.

How can we design a wearable system that provides 360-degree real-time object recognition and environmental risk alerts to enhance situational awareness and navigation safety?

Hypothesis

Suppose a wearable headset integrates real-time object detection (YOLO), speech feedback (TTS), and spatial awareness (LiDAR). In that case, users will be able to navigate environments more safely and efficiently—even under low-visibility conditions.

Ideas

  1. YOLO-based Object Detection System
  2. MobileNet SSD Model
  3. Dehazing Algorithm
  4. Text-to-Speech (TTS) and pyttsx3
  5. LiDAR Integration
  6. Configuration via JSON (Flag-based architecture)
  7. Multithreading & Frame Skipping

Comparative Study

Feature/Idea Strengths Weaknesses
YOLOv8 Accurate detection, well-suited for real-time Computationally heavy
MobileNet SSD Lightweight, faster Lower precision, not ideal for critical safety use
Dehazing Algorithm Enhances visibility in bad conditions Adds pre-processing step, slight latency increase
LiDAR + Risk Queue Enables distance-based alerts Complex to calibrate with camera system
pyttsx3 TTS Offline and customizable Less natural voice output than cloud-based TTS
JSON Config Architecture Modular and scalable Adds some complexity in state management
Multithreading & Skipping Improves efficiency and speed Increases debugging complexity, requires sync management

Approach

Screenshot 2025-07-30 233402.png

Results

<aside> ✅

YOLOv8 model delivered high object detection accuracy but needed performance optimization.

Screenshot 2025-07-25 164030.png

</aside>