Implementing Traffic Sign Recognition(TSR) by ELM

• An Efficient Method for Traffic Sign Recognition Based on Extreme Learning Machine
• Author: Zhiyong Huang, Yuanlong Yu, Jason Gu, and Huaping Liu
• Published: IEEE TRANSACTIONS ON CYBERNETICS, 2017.

TSR Architecture

The proposed method consists of two modules:

1. Extraction of histogram of oriented gradient variant (HoGv) feature
2. A single classifier trained by extreme learning machine (ELM) algorithm

How to Use it?

Step 1. Run the create_test_data_in_HoGv.m and create_train_data_in_HoGv.m to create the database of feature.

Step 2. Run the main.m to train the model and store the weighting of ELM (TSR_GTSRB.mat).

Step 3. Load the weighting and then inference.

Database

Opensource Database: GTSRB

The training data(39209 images,43classes) :

• Images and annotations (GTSRB_Final_Training_Images.zip)

The test dataset(12630 images) :

• Images and annotations (without ground truth classes) (GTSRB_Final_Test_Images.zip)
• Extended ground truth annotations (with classes) (GTSRB_Final_Test_GT.zip)

Notice

The dimenson of HoGv feature is fixed ,2500.

Experiment Based on GTSRB.

All experiments were carried out in a Matlab R2019a environment running on a desktop PC with a 3.8 GHz AMD Ryzen 5 3600X 6-Core Processor and a 16 GB memory.

• The Accuracy and time of training phase are 99.99% and 6.1ms/frame.
• The Accuracy of testing phase are 96.63%.