倒计时交通信号灯的检测与识别

Detection and Recognition of Countdown Traffic Lights

倒计时交通信号灯的检测与识别对于辅助驾驶和无人驾驶具有重要意义。基于深度学习的目标检测算法虽然是目前交通信号灯检测与识别领域的主流算法,但由于倒计时数字灯的读秒时间一般在100秒以上,单纯采用深度学习训练检测模型的方法难度较大,为此设计两种倒计时数字灯识别方法:第一种是基于传统图像处理与OCR相结合的方法,第二种是基于YOLOv5、图像处理、OCR相结合的方法,其中第二种方法表现较好,可有效提取数字灯ROI,并滤除噪声,在测试数据集上的识别准确率达95.6%。

The detection and recognition of countdown traffic lights are of great significance for both assisted driving and unmanned driving.Currently,Although the object detection algorithm based on deep learning is currently the mainstream algorithm in the field of traffic lights detection and recognition,However,due to the fact that the countdown time of digital lights is generally over 100 seconds,it is difficult to simply use deep learning to train the detection model,for this reason,designed two countdown digital light recognition methods:the first is based on a combination of traditional image processing and OCR,and the second is based on YOLOv5,image processing,and OCR.The second method performs well and can effectively extract the ROI of digital lights and filter out noise.with a recognition accuracy of 95.6%on the test dataset.