基于深度学习的道路交通目标自动化检测算法研究

Research on Road Traffic Target Automated Detection Algorithm Based on Deep Learning

在自动驾驶的研究中,道路交通目标的自动检测是最为关键的技术之一。然而目前的检测算法存在着漏检、误检等情况,严重危害到道路交通安全。为此,研究在深度学习基础上,添加了误检模块,并对检测算法的损失函数进行了改进,以提高检测的准确率。改进方法在三种难度的测试中,平均准确率达到了97.34%、87.24%和80.23%,相较于改进前分别提高了2.61%、3.01%和3.89%,并且优于目前较为先进的Faster R-CNN、Grid R-CNN和FreeAnchor算法。而在实际场景的测试中,改进后的算法在雨天和夜晚有效地减少了误检情况的发生。实验结果验证了此次研究的有效性,为道路交通目标自动化检测提供了有价值的参考。

In the research of autonomous driving,automatic detection of road traffic targets is one of the most critical technologies.However,current detection algorithms have situations such as missed and false detections,which seriously endanger road traffic safety.Therefore,on the basis of deep learning,a false detection module is added and the loss function of the detection algorithm is improved to improve the detection accuracy.The improved method achieved an average accuracy of 97.34%,87.24%,and 80.23%in three difficulty tests,which increased by 2.61%,3.01%,and 3.89%compared to before the improvement,and was superior to the currently more advanced Faster R-CNN,Grid R-CNN,and FreeAnchor algorithms.In actual scenario testing,the improved algorithm effectively reduces the occurrence of false positives on rainy days and nights.The experimental results validate the effectiveness of this study and provide valuable reference for automated detection of road traffic targets.