The fusion of the thermal target of infrared image and the clear background of visible image can realize the recognition of abnormal action associated with the scene under low illumination condition.The existing fusion methods which mainly focus on feature matching are affected by the differences of visible and infrared imaging scales,views,and target characteristics in monitoring scenes,and the efficiency and accuracy of registration and fusion are limited.To solve this problem,this paper proposes a fusion method of infrared and visible images from different views based on saliency detection.By presetting the heat-sensitive target,the transformation model of visible and infrared view is calculated,and the infrared and visible views are registered in advance.The saliency area of pedestrian target in infrared image is extracted with the Mask R-CNN network,and each target area is locally fused with visible image according to the view transformation model point.Finally,the identification of illegal intrusion action is verified by experiments.Experiments show that the fusion method proposed in this paper can effectively fuse the thermal target information of infrared image with the visible scene,and can accurately determine whether the illegal intrusion has occurred.