Cross-Modal Pedestrian Detection Algorithm Based on Gradient Convolution Fusion

Pedestrian detection is one of the important tasks in the field of computer vision. In recent years, the single-spectral pedestrian detection method widely used in visible light is easy to be affected by occlusion and illumination. In the night or dark light environment, it is easy to cause large deviation of detection results or miss detection. Firstly, the features of infrared and visible images are fused by using the Gradient Convolution Fusion Module (GCFM) designed in this paper, and then introduced into the improved YOLOv4 network for training and learning to achieve the purpose of pedestrian detection. In order to improve the detection accuracy, the attention mechanism is added to the YOLOv4 network. The proposed method is tested on the Kaist and MSRS datasets. The results show that the pedestrian detection algorithm proposed in this paper can improve the detection accuracy while ensuring real-time performance.