Pedestrian Detection Based on Improved Faster-RCNN Algorithm

In recent years, pedestrian detection based on image recognition has become an important research topic in vehicle assisted driving. For the question of poor detection accuracy resulted from missing detection and small targets in pedestrian detection, proposes a pedestrian detection method based on improved Faster-RCNN. First, ResNet34 residual network was used to replace VGG-16 as the backbone feature extraction network, and then SENet mechanism was introduced to further enhance and suppress the weight vector. Then, aiming at the multi-scale problem in the detection set, FPN network is added to further strengthen the feature extraction ability of the network. The k-means algorithm is introduced to generate appropriate anchors according to the characteristics of the dataset. The experimental results show that, compared with the classic network, the average precision (mAP) of the improved algorithm reaches 93.36%, which is 5.34% higher than the original Faster-RCNN algorithm, which proves the effectiveness of the algorithm.