Research on Pedestrian Detection Based on Improved SSD Algorithm

Pedestrian detection is an important part of assisted driving and automatic driving. This paper proposes a pedestrian detection method based on improved SSD (Single Shot MultiBox Dectector) algorithm for the problem of the detection accuracy in pedestrian test tasks. We reconstructed the feature extraction network of the SSD model. The MobileNet and ResNet (Residual Network) were used as the basic network of SSD to build the SSD model and improve the detection capability of the algorithm. We conducted experiments on public pedestrian detection data sets. The results were evaluated using mAP and FPS. The experimental results show that the mAP of the SSD algorithm after introducing the residual network reaches 85.00%, which is 11.92% higher than the original SSD algorithm, and the FPS reaches 39.31%, which effectively improves the accuracy and speed of pedestrian detection.