改进SSD算法在交通场景中的检测研究

针对目标检测算法SSD在交通应用中检测精度不高、对小尺度汽车和行人检测能力不足的问题,提出一种改进的SSD检测算法,将原SSD基础网络VGG-16替换成残差网络ResNet-50,来提高特征提取网络提取特征的能力并防止网络衰退.算法额外设计5层卷积层来简化原SSD网络结构,进行多尺度特征图的检测;将注意力机制CBAM融...

Research on Improved SSD Algorithm for Detection in Traffic

Aiming at the problems of low detection accuracy and insufficient detection ability of SSD in traffic applications, an improved SSD detection algorithm is proposed, which replaces the original SSD basic network VGG-16 with the residual network ResNet-50 to improve the feature extraction network’s ability to extract features and prevent the network from declining. The algorithm designs five additional convolution layers to simplify the original SSD network structure and detect multi-scale feature maps. The attention mechanism CBAM is integrated into the new basic network to improve semantic information extraction ability, small target detection ability and overall network accuracy. Experimental results show that the detection accuracy of the improved algorithm on KITTI data set reaches 62.0%, which is 4.7%higher than that of the original SSD network.