Traffic sign recognition system using deep learning.
Abstract
Traffic signs are an essential part of any transport system, and failure of detection by the
driver may significantly increase the accident risk. Traffic sign detection also plays an
important role in autonomous driving. In this report we design a traffic sign recognition
system by applying deep learning techniques, a subfield of AI and in particular using the SSD
algorithm. Compared with other detection systems, this system offers real-time detection
and thus can be incorporated in a larger driver assistance system. The objective is locating
and classifying traffic signs in natural street scenes, drawing a bounding box around it,
identifying the type of traffic sign and relaying this information to the driver dashboard.
The machine learning model developed is trained on Ugandan traffic signs collected locally
and is thus applicable on Ugandan roads. The test results show that the model accurately
locates the traffic signs confirming its robustness and suitability in real-world applications.
The system detects only a subset of the Ugandan traffic signs considering the shortage of
data for some traffic signs. This will however lay a foundation for subsequent research in
traffic sign recognition systems. Traditional Evaluation parameters: mAP(mean Average
precision-Precision, Recall and IoU) and FPS(Frames per second) are run-down to analyze
the accuracy and speed of the two SSD algorithms trained in this report i.e. one built with
a ResNet backbone and one with a MobileNet backbone.