Scholarlink

<p>Traffic sign recognition is essential for contemporary autonomous driving systems and intelligent transportation infrastructure. Nevertheless, the majority of current traffic sign recognition models analyze frames individually and lack reliability in real world settings, including low lighting, occlusion, and cluttered backdrops, which are particularly prevalent on Indian roads. This research presents the Temporal Integrated CNN (TI-CNN), an innovative lightweight deep learning system developed for real time video based traffic sign detection. In contrast to traditional traffic sign identification models, TI-CNN employs a Temporal Consistency Module (TCM) to assimilate temporal information from consecutive video frames, hence enhancing detection stability. The novelty of the TI-CNN lies in its utilization of the TCM (Temporal Convolutional Model), which not only ensures that the final result of a detected object will remain stable over time, but it is also able to deal with a common issue that occurs with driving video footage, such as flickering results of detections, as exhibited by the circumstances with the driving video images taken from Indian highways. Thus, through both the use of and integration of the TCM, the TI-CNN is able to be distinguished from all other CNNs (Convolutional Neural Networks) and be given additional robustness in terms of interference caused by other occluding objects, environmental lighting conditions, as well as blur resulting from moving objects in the environment. Therefore, by using a combination of GTSRB, GTSDB, BTSD benchmark datasets, and a custom dataset specifically generated for Indian traffic signs, containing four categories of signs (speed limit signs, turn signs, zone signs and bump signs), performance improved and accuracy increased to 91.4%, with a high degree of temporal stability (92.3%) at 28.5 frames per second (FPS), which indicates that the TI-CNN appears to have significant potential to perform in practical environments, such as those that exist within India where complex and varying situations are common.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Temporal integrated CNN for robust real time video based traffic sign recognition on indian roads

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Kathiresan Kandasamy,
  • Yuvaraj Natarajan,
  • Sri Preethaa K. R.,
  • Sathish Kumar G.

摘要

Traffic sign recognition is essential for contemporary autonomous driving systems and intelligent transportation infrastructure. Nevertheless, the majority of current traffic sign recognition models analyze frames individually and lack reliability in real world settings, including low lighting, occlusion, and cluttered backdrops, which are particularly prevalent on Indian roads. This research presents the Temporal Integrated CNN (TI-CNN), an innovative lightweight deep learning system developed for real time video based traffic sign detection. In contrast to traditional traffic sign identification models, TI-CNN employs a Temporal Consistency Module (TCM) to assimilate temporal information from consecutive video frames, hence enhancing detection stability. The novelty of the TI-CNN lies in its utilization of the TCM (Temporal Convolutional Model), which not only ensures that the final result of a detected object will remain stable over time, but it is also able to deal with a common issue that occurs with driving video footage, such as flickering results of detections, as exhibited by the circumstances with the driving video images taken from Indian highways. Thus, through both the use of and integration of the TCM, the TI-CNN is able to be distinguished from all other CNNs (Convolutional Neural Networks) and be given additional robustness in terms of interference caused by other occluding objects, environmental lighting conditions, as well as blur resulting from moving objects in the environment. Therefore, by using a combination of GTSRB, GTSDB, BTSD benchmark datasets, and a custom dataset specifically generated for Indian traffic signs, containing four categories of signs (speed limit signs, turn signs, zone signs and bump signs), performance improved and accuracy increased to 91.4%, with a high degree of temporal stability (92.3%) at 28.5 frames per second (FPS), which indicates that the TI-CNN appears to have significant potential to perform in practical environments, such as those that exist within India where complex and varying situations are common.