Pedestrian detection based on attention mechanism and feature enhancement with SSD

Some factors such as low resolution of small targets, limited target features, and noise interference affect the effect of pedestrian detection. Therefore, a pedestrian detection algorithm based on attention mechanism and feature enhancement with SSD is presented in this paper. It uses channel feature fusion to fuse non-adjacent convolutional layers to obtain significant edge gradient features and semantic information features. Finally, through the optimization of the attention mechanism CBAM, the channel features and spatial features are coupled under different fusion detection layers to improve the feature weight of the pedestrian's salient region, so as to the detection accuracy of the algorithm has taken a big step forward. The improved algorithm is verified on the fusion dataset and the VOC2007TEST dataset, compared with the SSD algorithm, the detection accuracy of the improved algorithm model on the fusion dataset reaches 60.1%, with an increase of 7.2%, and that on the VOC2007TEST dataset reaches 78%, with an increase of 4.1%. The experimental results show that this method can effectively detect the small target pedestrian in the image, and reduce the error detection and omission detection, thus verifying its feasibility.