Deep traffic light detection by overlaying synthetic context on arbitrary natural images

Abstract

Deep neural networks come as an effective solution to many problems associated with autonomous driving. By providing real image samples with traffic context to the network, the model learns to detect and classify elements of interest, such as pedestrians, traffic signs, and traffic lights. However, acquiring and annotating real data can be extremely costly in terms of time and effort. In this context, we propose a method to generate artificial traffic-related training data for deep traffic light detectors. This data is generated using basic non-realistic computer graphics to blend fake traffic scenes on top of arbitrary image backgrounds that are not related to the traffic domain. Thus, a large amount of training data can be generated without annotation efforts. Furthermore, it also tackles the intrinsic data imbalance problem in traffic light datasets, caused mainly by the low amount of samples of the yellow state. Experiments show that it is possible to achieve results comparable to those obtained with real training data from the problem domain, yielding an average mAP and an average F1-score which are each nearly 4 p.p. higher than the respective metrics obtained with a real-world reference model.

An Augmented Reality Periscope for Submarines with Extended Visual Classification

“…Still, this problem faces similar issues with detecting cars or traffic signs using a camera in autonomous or semi-autonomous vehicles. De Mello et al [10] has proposed a method to generate artificial traffic-related training data for deep traffic light detectors, offering a solution using deep neural networks for problems associated with autonomous driving. Concerning vessel detection and classification, Kim et al [11] proposed a novel probabilistic ship detection and classification system based on deep learning using a dataset of images available on the web.…”

Traffic lights detection and tracking for HD map creation

“…Similarly, the DriveU Traffic Light Dataset (DTLD) by Fregin et al (2018a) contains traffic lights with a width of 1-88 pixels and a mean of 14.67 pixels. Many studies have focused on traffic light detection, and state recognition, like , and (Possatti et al, 2019), or (de Mello et al, 2021. These studies perform well and detect tiny traffic light objects in the input images.…”

A new traffic signal detection system with high robustness and real-time requirements

“…The traffic light can be detected from overexposed and underexposed images with a robust image acquisition method 2 . A light-weight deep CNN (Convolutional Neural Network) model and RM R-CNN (Refined Mask R-CNN) model are introduced for the automatic traffic light recognition [3][4][5][6] . The support vector machine (SVM) and histogram of oriented gradients (HOG) features are utilized to detect the state of traffic light 7 .…”