TrafficSignDetection : Machine Learning Model to Detect Road Signs
This is an introduction to「TrafficSignDetection」, a machine learning model that can be used with ailia SDK. You can easily use this model to create AI applications using ailia SDK as well as many other ready-to-use ailia MODELS.
Overview
TrafficSignDetection is a machine learning model for detecting road signs released in November 2018.
Architecture
TrafficSignDetection uses one Faster R-CNN, R-FCN, SSD or YOLOv2 state-of-the-art object-detection model architecture, pre-trained on the Microsoft COCO dataset and then fine-tuned on the German Traffic Sign Detection Benchmark (GTSDB) dataset.
The mAP using Faster R-CNN ResNet50 is 91.52.
Three categories can be detected: prohibitory, mandatory, and danger.
Although it was trained on a dataset made of German road signs, many of the designs are common to other countries and the model can also be used in those countries. The capture below shows the result on a street in Japan.
Below are some examples of German traffic signs included in the GTSRB dataset.
The following chart shows the main signs used in Japan. Designs for speed limits and one-way streets for example are similar to Germany and the model can detect them just fine.
Usage
TrafficSignDetection can be used with ailia SDK 1.2.10 using the following command.
$ python3 traffic-sign-detection.py --input input.jpg --savepath output.jpg
ax Inc. has developed ailia SDK, which enables cross-platform, GPU-based rapid inference.
ax Inc. provides a wide range of services from consulting and model creation, to the development of AI-based applications and SDKs. Feel free to contact us for any inquiry.