Research Article Transportation Research Record 1–9 Ó National Academy of Sciences: Transportation Research Board 2020 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0361198120935451 journals.sagepub.com/home/trr Autonomous Vehicle Application for Improving Traffic Sign Learning near Ramps Zhenhua Zhang 1 , Leon Stenneth 1 , and Xiyuan Liu 2 Abstract The state-of-the-art traffic sign recognition (TSR) algorithms are designed to recognize the textual information of a traffic sign at over 95% accuracy. Even though, they are still not ready for complex roadworks near ramps. In real-world applications, when the vehicles are running on the freeway, they may misdetect the traffic signs for the ramp, which will become inaccurate feedback to the autonomous driving applications and result in unexpected speed reduction. The misdetection problems have drawn minimal attention in recent TSR studies. In this paper, it is proposed that the existing TSR studies should transform from the point-based sign recognition to path-based sign learning. In the proposed pipeline, the confidence of the TSR obser- vations from normal vehicles can be increased by clustering and location adjustment. A supervised learning model is employed to classify the clustered learned signs and complement their path information. Test drives are conducted in 12 European countries to calibrate the models and validate the path information of the learned sign. After model implementa- tion, the path accuracy over 1,000 learned signs can be increased from 75.04% to 89.80%. This study proves the necessity of the path-based TSR studies near freeway ramps and the proposed pipeline demonstrates a good utility and broad applicability for sensor-based autonomous vehicle applications. The traffic sign recognition (TSR) system can detect and track road signs and display the sign information in the vehicle. TSR applications play an increasingly important role in driving assistance and even the autonomous driving in today’s highly regulated road environment. Over the past years, the image processing technologies have matured and achieved a fairly high accuracy in TSR appli- cations (1–3). However, most of the models are tested using some publicly available data sets, and consequently they are inevitably not able to cover all the challenges in real driving environments. Also, in considerations of reducing costs and increasing reliability, TSR performance may be much more conservative when put into commer- cial use and thus the corresponding accuracy may not be as high as that under experimental settings. For instance, Figure 1a shows a field image by a TSR system in a used car which is highly blurred and the detection result shows a wrong detection of a 90 km/h speed limit. The TSR application is sometimes not reliable and can be influenced by factors such as the industry design, year of service, ver- sion of TSR system, and so forth. Besides the equipment wear and image classification errors, the accuracy of TSR applications will also be greatly reduced near ramps, which is the major concern in this paper. Signs may be accurately detected but applied on the wrong links. For instance, in Figure 1b, TSR system can detect a speed limit sign of 40km/h on the freeway which should be applied on the ramp. Without proper calibration, autonomous vehicles will get feeds from this TSR system and adjust the speed accordingly. Consequently, there is an unexpected and uncomfortable speed reduction on the freeway. The same problems may happen in Figure 1c where vehicles either on the main link or on the ramp cannot decide the proper speed limit value. To improve the TSR near complex road geometries, this paper proposes a systematic pipeline to increase the confidence of the TSR observations and complement their path information. The focus is on on the speed limit signs near the ramps and the aim is to ready them for autonomous vehicle applications. In this pipeline, the TSR aggregation of nearby vehicles will deal with the 1 HERE Technologies, Chicago, IL 2 University at Albany – State University of New York, Albany, NY Corresponding Author: Zhenhua Zhang, zhenhuaz@buffalo.edu