search

Akasxh / Terrain-Recognition

High accuracy, explainable, lightweight CNN for terrain recognition.
GNU General Public License v3.0
11 stars 16 forks source link

Hybrid Terrain Classification with Edge Computing #6

Open Shibika-Roy opened 1 month ago

Shibika-Roy commented 1 month ago

Overview This proposal outlines the development of a cutting-edge open-source hybrid terrain recognition system that combines image-based terrain classification with sensor data (such as accelerometers, gyroscopes, and soil sensors) for enhanced accuracy and robustness. The project will also focus on optimizing the system for edge computing, making it capable of running efficiently on low-power, resource-constrained devices like drones, robots, and mobile devices.

The goal is to create a solution that can classify terrains with higher precision by leveraging both visual input and physical measurements, ensuring real-time performance in environments with limited computational resources and connectivity.

Model Description: Current terrain classification models, primarily relying on image recognition, face challenges in real-world applications where visuals may be unclear due to factors like poor lighting, weather conditions, or overlapping terrain features. Additionally, image-only models are often resource-intensive, making them impractical for deployment on edge devices.

The proposed hybrid system would enhance classification accuracy and resilience by fusing image data with sensor inputs. Moreover, by utilizing edge computing techniques, this solution can be applied in scenarios such as disaster response, autonomous navigation, and environmental monitoring, where real-time local processing is essential.

Proposed Features:

  1. Image-Based Terrain Classification (CNN)
  2. Sensor Fusion
  3. Edge Computing Optimization
  4. Real-Time Terrain Classification

As a contributor, I will take responsibility for:

Designing and implementing the CNN model for image-based terrain recognition. Curating a robust dataset and managing the integration of additional sensor data. Collaborating on edge computing optimizations to ensure the model can run on low-resource devices. Actively engaging with the community to gather feedback and make iterative improvements to the project. Please kindly assign this issue to me and label it as gssoc24-extd and hacktoberfest. I'm looking forward to it .

Akasxh commented 1 month ago

I have assigned this issue to you. I would assign a suitable label on completion of your project.

This has around 10k images for you to work around: https://drive.google.com/drive/folders/1hbL1m39TF8ABe0oCj5XYDbHXY-gPIjcQ?usp=drive_link

While curating your dataset, make sure to contribute the zip file of overall dataset for others to work on.

Thank you

AKSHITHA-CHILUKA commented 1 month ago

@Akasxh despite of many times telling you have not changed the labels. The crt labels are level3 with no spaces , gssoc-ext is the label not gssoc24-extd , hactoberfest-accepted is the label not hactoberfest . I hope you will change all the label's in all the issues and pr's as soon as possible .