The Dataset I have used here is the German Traffic Sign Benchmark is a multi-class, single-image classification challenge held at the International Joint Conference on Neural Networks (IJCNN) 2011. Traffic sign detection is a high relevance computer vision problem and is the basis for a lot of applications in industry such as Automotive etc. Traffic signs can provide a wide range of variations between classes in terms of color, shape, and the presence of pictograms or text. In this project, I used deep neural networks and two classic convolutional neural network architectures (LeNet and AlexNet) to classify traffic signs. I will train and validate a model so it can classify traffic sign images using the German Traffic Sign Dataset. After the model is trained, I will then try out my model on images of German traffic signs that I find on the web.
The motivation for this project lays both personal interest in a better understanding for object detection and academic research. The goal is to develop a foundation for a roadsign-detection (RSD) with the option to add further objects or functions to it. The ultimate goal is to have useable object detection for the automotive sector.
The goals / steps of this project are the following:
- Load and explore the data set.
- Realize LeNet architecture and use ReLu, mini-batch gradient descent and dropout.
- Realize AlexNet and make some modifications, use learning rate decay, Adam optimization and L2 regulization.
- Analyze the softmax probabilities of the new images
- Summarize the results