Overview / Usage

• Harnessed the power of Intel technologies to enhance the capabilities of the YOLOv5 algorithm for object detection. By leveraging Intel's optimized libraries and frameworks, such as Intel oneDAL, Intel optimized PyTorch, and the SYCL/DPC++ libraries, we have achieved superior performance, accuracy, and efficiency in our object detection model. This integration enables us to process data faster, optimize resource utilization, and streamline post-processing steps, leading to robust and real-time object detection for autonomous vehicle applications.
• The purpose of our project is to leverage Intel technologies to enhance the YOLOv5 algorithm for object detection in the context of autonomous vehicles. By utilizing Intel oneDAL, Intel optimized PyTorch, and the SYCL/DPC++ libraries, we aim to achieve improved performance, accuracy, and efficiency in detecting and classifying objects in real-time. Our goal is to provide a reliable and effective solution for autonomous vehicles to detect and respond to various objects and obstacles on the road, ensuring enhanced safety and efficiency in autonomous driving systems.

Methodology / Approach

Our approach to solving the problem of object detection for autonomous vehicles involves harnessing the power of Intel technologies to enhance the capabilities of the YOLOv5 algorithm. To achieve this, we utilize several Intel-optimized libraries, frameworks, and techniques, ensuring superior performance, accuracy, and efficiency.

• Intel Technologies and Frameworks: We leverage Intel oneDAL, Intel-optimized PyTorch, and the SYCL/DPC++ libraries as our primary tools. These frameworks enable us to process data faster, optimize resource utilization, and streamline post-processing steps. The Intel AI Analytics Toolkit, featuring optimized deep learning frameworks like PyTorch and TensorFlow, powers our prototype.
• Model Creation: Our journey begins with the creation of a highly efficient object detection model. We start by curating a labeled dataset, which serves as the foundation for training our model. Through the ETL (Extract, Transform, Load) process, we preprocess and clean the data to ensure its quality and readiness for model training.
• Model Training: The heart of our solution lies in training the deep learning model. We utilize Intel-optimized libraries like oneDNN and oneDAL to train and infer deep learning models. These libraries are instrumental in optimizing the training process, leading to better model performance.
• Real-time Object Detection: Our ultimate goal is to provide real-time object detection capabilities for autonomous vehicles. Leveraging the YOLO (You Only Look Once) object detection algorithm, our model can quickly and accurately identify and classify objects in the vehicle's surroundings.
• Deployment: Once the model is trained and fine-tuned, we export it for deployment in autonomous vehicles. Our system integrates seamlessly with the vehicle's sensors and cameras, processing the video feed in real-time.
• Enhancing Autonomous Driving: Our solution enhances the safety and efficiency of self-driving cars. By leveraging Intel technologies and frameworks, we've created a robust system that combines advanced computer vision algorithms and deep learning models. It can identify objects, detect traffic signals, and make decisions in real-time to ensure safe and efficient autonomous driving.

Technologies Used

Technologies:

1. Intel® Distribution for Python: A Python distribution optimized for high-performance computing, enabling efficient code execution.

Libraries:

1. OpenCV (opencv-python=4.5.5.64): A powerful computer vision library used for various image and video processing tasks, including object detection.
2. NumPy (numpy=1.23.4): A fundamental package for scientific computing in Python, crucial for array operations and data manipulation.
3. PyTorch (pytorch=1.13.0): A deep learning framework used for implementing the YOLOv5 algorithm, facilitating efficient object detection.
4. Intel® Extension for PyTorch (intel-extension-for-pytorch=1.13.0): A package that enhances PyTorch's performance on Intel architectures, optimizing the execution of PyTorch-based models.

Intel Technologies:

1. Intel® Neural Compressor (neural-compressor=2.0): A technology used for neural network compression, potentially reducing the computational demands of deep learning models.
2. Intel® Distribution of OpenVINO™ (openvino-dev=2022.3.0): An Intel distribution designed for optimized deep learning inference. It enables efficient deployment of deep learning models on Intel hardware, such as CPUs, GPUs, and accelerators.

Hardware: The specific hardware used in this project may vary, but Intel technologies are optimized to work on a range of Intel CPUs, GPUs, and other hardware accelerators, ensuring efficient execution and high-performance inference.

Repository

https://github.com/Senthil-Riddhish/Intel-One-Api-Hackathon

Other links

• Intel® oneAPI medium article