On Salience-Sensitive Sign Classification in Autonomous Vehicle Path Planning: Experimental Explorations with a Novel Dataset

Abstract

One of the most important tasks for ensuring safe autonomous driving systems is accurately detecting road traffic lights and accurately determining how they impact the driver's actions. In various real-world driving situations, a scene may have numerous traffic lights with varying levels of relevance to the driver, and thus, distinguishing and detecting the lights that are relevant to the driver and influence the driver's actions is a critical safety task. This paper proposes a traffic light detection model which focuses on this task by first defining salient lights as the lights that affect the driver's future decisions. We then use this salience property to construct the LAVA Salient Lights Dataset, the first US traffic light dataset with an annotated salience property. Subsequently, we train a Deformable DETR object detection transformer model using Salience-Sensitive Focal Loss to emphasize stronger performance on salient traffic lights, showing that a model trained with this loss function has stronger recall than one trained without.

Trajectory Prediction in Autonomous Driving With a Lane Heading Auxiliary Loss

“…However, this is not the only way to cluster such data, nor is this the only possible data which can be used in the clustering process; future research can iterate on these methods to further drive development of learning systems which select data at low cost to human annotators and intelligently guide data curation at scale. While we provide a selection based on "novelty" or "uniqueness" in this research, other measures, such as salience [27], [28], [29] or even language-based queries [30], may also be highly informative to efficient and safe model learning [31]. Beyond learning itself, such novelty-mining is also important in selection of data for system validation [32].…”

Robust Perception and Visual Understanding of Traffic Signs in the Wild

“…Several studies have shown that visual features such as color, contrast, shape complexity, and size, can influence the salience and detectability of traffic signs, but algorithms that estimate sign relevance are limited. For instance, Greer et al [16] conducted a preliminary study on traffic sign salience recognition and introduced a novel dataset that considers sign salience. Their CNN predicts the sign salience with 76% accuracy.…”

Real-Time Traffic Sign Detection: A Case Study in a Santa Clara Suburban Neighborhood

“…In the future, such algorithms could be paired with autonomous vehicles to assist driving capabilities and prompt maneuvering to prevent a collision in response to road or traffic conditions. [14] This work aims to generate a suitable algorithm to detect real-time traffic signage and test its efficacy and sensitivity. The approach to this work was completed using the YOLOv5 computer vision algorithm.…”