To solve the problem that the Tiny YOLOv4 target detection algorithm has low accuracy and low recall rate in pedestrian detection, the feature extraction network and prediction network are improved. In the part of feature extraction network, the traditional convolution network is replaced by a depthwise separable convolution network to reduce parameters and computation. The attention mechanism module is added in the feature extraction network to enhance the area of interest of detecting object and improve the detection accuracy. A prediction scale is added in the prediction network, and the added scale is enhanced by features to improve the recall of detection of objects. The experimental results show that compared with the original algorithm, the improved Tiny YOLOv4 algorithm improves the accuracy by 7.1%, and the recall rate also increases by 6.6%.