Traffic Sign Detection Under Challenging Conditions: A Deeper Look into Performance Variations and Spectral Characteristics

Abstract

Traffic signs are critical for maintaining the safety and efficiency of our roads. Therefore, we need to carefully assess the capabilities and limitations of automated traffic sign detection systems. Existing traffic sign datasets are limited in terms of type and severity of challenging conditions. Metadata corresponding to these conditions are unavailable and it is not possible to investigate the effect of a single factor because of simultaneous changes in numerous conditions. To overcome the shortcomings in existing datasets, we introduced the CURE-TSD-Real dataset, which is based on simulated challenging conditions that correspond to adversaries that can occur in real-world environments and systems. We test the performance of two benchmark algorithms and show that severe conditions can result in an average performance degradation of 29% in precision and 68% in recall. We investigate the effect of challenging conditions through spectral analysis and show that challenging conditions can lead to distinct magnitude spectrum characteristics. Moreover, we show that mean magnitude spectrum of changes in video sequences under challenging conditions can be an indicator of detection performance. CURE-TSD-Real dataset is available online at https://github.com/olivesgatech/CURE-TSD.

Survey and Performance Analysis of Object Detection in Challenging Environments

“…To increase the detection of small traffic signs, the authors used the L1 constraint term to modify Tversky Loss [98] instead of the conventional intersection over union loss. To appraise the system, the author used the CURE-TSD dataset [99] and achieved a precision of 94.60% and recall of 80.21% beating previous state-of-the-art results.…”

“…Given five different levels of challenges to each type of occlusions, there is a total of 1.72 million frames. There are 14 types of traffic signs in this dataset labeled as speed limit, goods vehicles, no overtaking, no stopping, no parking, [106] explaining harsh weather, (c) describes samples from CURE-TSD [99] illustrating example of camera distortion, lens flare, (d) denotes samples from KITTI [96] depicting various objects to be detected, (e) represents samples from Kaist [68] explaining saliency maps with their night time images, (f) depicts samples taken from UNIRI-TID [61] showing example of thermal images, (g) highlights samples from SKU-110K [76] representing example of cluttered objects, (h) represents samples taken from Wider Face [71] showing faces at various angles, (i) represents samples taken from VOT-2018 [103] presenting example of complex indoor scenes, (j) shows samples taken from DFG [89] illustrating traffic signs at various places, (k) represents images taken from MS-COCO [12] describing example of objects in daily life and (l) outlines samples taken from See in the dark [59] dataset exhibiting examples captured at low illumination and high exposure.…”

“…Figure 15 (b) depicts few samples from this dataset. We incorporate this dataset along with ExDARK [105] and CURE-TSD [99] to benchmark the results on state-of-the-art object detection algorithms.…”

Sensors on the Move: Onboard Camera-Based Real-Time Traffic Alerts Paving the Way for Cooperative Roads

“…Currently, the improvement is focused on solving both detection and classification phases together in one step [16]. Additionally, researchers are also trying to solve this issue in all kind of weather and light situations [17,18].…”

“…This has allowed many comparative studies that have helped to improve existing algorithms. However, most of the existing datasets reached saturation, since most of the results are in the range 95-99% of the perfect solution [22,23] Nevertheless, external non-technical challenges, such as lighting variations and weather condition changes, occlusions or damaged images, even variations in traffic signs among different countries, may decrease the system's performance [17].…”