Research on traffic sign detection algorithm based on deep learning

摘要

Due to the respective interests among themain bodies of the supply chain, it is necessary to introducesome mechanisms to coordinate the problem of decrease of detection rate caused by theinterconnection of closed-loop traffic signs. This paper proposes a traffic sign detection algorithmbasedondeep learning. It uses the red, green, andblue (RGB)normalization-based colordetectionalgorithm and regional feature decision criteria to automatically identify themulti-sign interconnectioncandidate regions and perform edge smoothing and contour tracking for the extractedtarget regions. It uses deep learner based on global and local curvature characteristics to maketraffic sign detection on the extracted contours, and, according to judgment criteria of convexityand concavity of corners as well as matching conditions of detection point pairs, extracts thedetection point pairs between the signs from the corners. It seeks the detection lines betweendetection point pairs and realizes the final detection of signs. The experimental results verify theeffectiveness of the proposed algorithm. Compared with the existing sign detection algorithmbased on watershed transformation and the improved adaptive detection algorithm, it overcomesthe sign over-detection problem and improves the overall performances of the sign detection.