Overview / Usage

One of the applications where deep learning is being used extensively is Autonomous Driving. The autonomous driving mechanism needs to detect innumerable objects (pedestrians, other cars, obstacles, etc) and make decisions. One of the smaller problems is detecting traffic signs and making decisions accordingly. I'm going to solve problems of detecting traffic signs on the road and detecting vehicles.

Methodology / Approach

To develop this solution I have used the deep learning algorithms of a Convolutional Neural Network that has multiple layers deployed in a U-net Architecture. Architecture is employed using a training dataset containing around 39,000 images, while the test dataset contains around 12,000 images containing 43 different classes. I will be using Convolutional Neural Networks(CNN) to solve this problem using the Keras framework and TensorFlow as the backend.

From the given video footage, moving objects are detected. The object detection model Single Shot Detector is used to classify those moving objects into their respective classes. It improves accuracy with many tricks and is more capable of detecting objects. The classifier model is built with Darknet-53 architecture. Table-1 shows how the neural network architecture is designed.

Using the optimized libraries (oneDNN) and (oneDAL) of oneAPI in TensorFlow 11.0 and sklearnex versions respectively played a significant role in achieving the required training output in just 35 minutes, making it the most notable tool among all. This optimization significantly boosted the speed of our training process. We are also exploring the possibility of further enhancing the speed by using the OpenVino toolkit, which has the potential to reduce the training time to just 20 minutes. The training time required by the PC was almost twice as long as the time taken by the oneAPI toolkit in DevCloud.

Technologies Used

Python, CNN, oneDAL, oneDNN

Repository

https://github.com/Manikanta-7342/SafeStreet