Traffic Sign Identification

Traffic signs are utilized as a method of warning and guiding drivers, helping to regulate the flow of traffic among vehicles, pedestrians, motorcycles, bicycles, and others who travel the streets, highways, and other roadways. Traffic sign detection is a crucial task in autonomous driving systems.

Project Title

Traffic Sign Identification

Project Area of Specialization

Artificial Intelligence

Project Summary

Traffic signs are utilized as a method of warning and guiding drivers, helping to regulate the flow of traffic among vehicles, pedestrians, motorcycles, bicycles, and others who travel the streets, highways, and other roadways. Traffic sign detection is a crucial task in autonomous driving systems. Due to its importance, several techniques are proposed to solve this problem, however, the problem still persists due to a number of associated issues, for example, occlusion of objects in the image, blurry signs, motion artifacts in the image due to vehicle movement etc. To overcome these problems, this works conceptualizes a robust solution that improves the segmentation of the traffic sign and improves the detection process. The conceptual novelty of this work lies in the development of a constrained size, cascaded deep learning model that will operate on traffic sign images to precisely detect the traffic signs. The proposed solution will be benchmarked on a standardized traffic signs dataset i.e., the German Traffic Sign Recognition Benchmark (GTSRB) dataset. Additionally, we also intend to test the proposed system for Pakistani traffic signs. Once developed, it will be helpful for drivers while driving i.e., in case a sign is missed by the driver, the system will notify the driver without having them to shift their focus.

Project Objectives

1. A system capable of detecting traffic signs using traffic sign images.
2. The system will try to detect signs having a number of associated issues, for example, occlusion of objects in the input image, blurry signs, motion artifacts in the image due to vehicle movement etc.

Project Implementation Method

The proposed solution for traffic sign detection mainly consists of four steps, namely: 1) Sign Pre-Processing, 2) Geometrical Maps of signs, 3) Cascaded Model, and 4) Weighted Projections.

1) Sign Pre-Processing: To perform these steps, in the first part, the signs will be pre-processed to remove noise. This step will involve intensive noise modeling of a cluster of signs to identify the type of noise present in the traffic sign images and remove it. All the images in the second stage will be down sampled to a lower resolution for example 64x64 dimension grid. This down sampled image will be the input of the cascaded model.

2) Geometrical Maps of signs: The geometrical maps will be constructed from raw traffic sign images and will contain the geometrical attributes of the objects. These maps will be second input of the cascaded model.

3) Cascaded Model: The proposed cascaded model will consist of two UNET architectures each operating on upper and lower portion of the processed X-ray image, along with their geometrical maps. The UNET architecture outperforms other models in segmentation tasks, however, relatively underperforms in segmenting noisy regions of the images. The conceptual novelty of this work lies in designing robust up sampling, and down sampling filters that improve the overall segmentation. The output of the deep learning model will be projected with weighted kernels.

4) Weighted Projections: Mostly, since the prevalence of signs varies in geometrical maps and noiseless images, therefore, we will separately process both parts in a cascaded UNET architecture. Moreover, we assign separate weights to both parts, in order to control the influence of one over the other. These projections will be learned and will output the probability of each traffic sign in the input image. The proposed solution will be benchmarked on a standardized traffic signs dataset i.e., German Traffic Sign Recognition Benchmark (GTSRB) dataset. Additionally, we also intend to test the proposed system for Pakistani traffic signs. Once developed, it will be helpful for drivers while driving i.e., in case a sign is missed by the driver, the system will notify the driver without having them to shift their focus.

Benefits of the Project

1) This framework can be used automated vehicles for sign identification and choosing an action accordingly.

2) It can also be used in gadgets of people with special needs to help them identify the traffic signs and to make them aware of basic traffic laws, along with giving warnings and information about locations and nearby amenities.

Technical Details of Final Deliverable

In this project we are going to develop a deep learning based framework to classify road signs using images. The following are the project deliverables:

1. A comprehensive report that will help to reproduce the project.

2. A software + hardware system that will detect the traffic signs in real-time.

3. A trained model on the German Traffic Sign Recognition Benchmark (GTSRB) dataset.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

IT

Other Industries

Transportation

Core Technology

Artificial Intelligence(AI)

Other Technologies

Others

Sustainable Development Goals

Good Health and Well-Being for People, Industry, Innovation and Infrastructure, Life on Land

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com