一种基于特征融合的道路目标检测方法

A Road Target Detection Method Based on Feature Fusion

当前复杂的道路交通环境给公众的安全出行带来了严峻挑战。在实际道路交通场景中进行目标检测时,传统的目标检测方法难以提取到能够适用道路多目标检测的有效特征,进而影响目标检测的准确性。文章以道路目标精准检测为研究出发点,结合当前先进的卷积神经网络技术,提出了一种基于特征融合的目标检测方法,即采用多个不同尺度的卷积核对特征提取方法进行改进,并通过所设计的特征融合及优化方案对所提取的特征进行更深入的表达。仿真实验证明,新方法与原始Faster-RCNN方法相比,对目标车辆的检测准确率提升了0.06%,对目标行人的检测准确率提升了0.16%,为复杂道路交通环境下的道路目标检测提供了可靠的实现方法与途径。

The current complex road traffic environment brought serious challenges to the safe travel for the public.Facing the problem of target detection in actual road traffic scenes,it is difficult to use the traditional target detection methods to extract effective features suitable for multi-target detection on the roads,which affects the accuracy of target detection.Starting from the basis of accurate road target detection and combining the current advanced convolutional neural network technology,this paper proposes a target detection method based on feature fusion,that is,multiple convolutional kernels with different scales are used to improve the feature extraction methods,and the extracted features are further expressed through the designed feature fusion and optimization schemes.Through simulation experiments,the detection accuracy of target vehicles and target pedestrians is improved by 0.06%and 0.16%respectively compared with the original Faster-RCNN method,which makes it reliable for implementing the road target detection in complex road traffic environment.