Overview / Usage

Overview:
This project uses deep learning to automatically detect accidents within uploaded video clips, addressing the challenge of unnoticed road incidents.

Problem:
Manually monitoring vast amounts of CCTV footage for accidents is inefficient. Our solution automates this, ensuring accidents are promptly identified.

Application:
Ideal for traffic systems, security setups, and autonomous vehicles, this tool ensures timely accident recognition and response.

Methodology / Approach

Methodology: We employed a ResNet-50 model, initially trained for 5 epochs, then optimized using Intel's IPEX for 15 epochs. Post-training, it was converted to ONNX and further optimized with OpenVINO. The streamlined model powers our Gradio app, allowing users to upload videos for real-time accident detection.

Technologies Used

Technologies & Libraries:

1. Deep Learning Framework: PyTorch
2. Model Architecture: ResNet-50
3. Model Optimization: IPEX, OpenVINO
4. App Interface: Gradio
5. Image Processing: OpenCV

Tools & Software:

1. ONNX (for model conversion)
2. Gradio (for creating a web interface for model deployment)

Hardware & Cloud Environment:

• Intel DevCloud: Utilized for the entire training process, leveraging its advanced computation capabilities.

Intel Technologies:

1. IPEX: Used for optimizing the model post initial training.
2. OpenVINO: Employed to further optimize the ONNX-converted model, ensuring faster inference speeds.

The training, optimization, and conversion processes were all executed seamlessly on the Intel DevCloud, ensuring a streamlined development workflow.

Repository

https://github.com/SSKlearns/IntelOneAPI/