Deep learning based nighttime target enhancement detection algorithm for intelligent vehicles

Abstract

Nighttime environment perception is a difficult problem for intelligent vehicles technology research. Although unimodal infrared images and visible images are widely used in the field of target detection, they are difficult to meet the needs of nighttime road target detection, so this paper proposes a deep learning based nighttime target enhancement detection algorithm for intelligent vehicles from the perspective of bimodal fusion. First, the input images are preprocessed using various infrared-visible bimodal image fusion algorithms, and four parameters such as standard deviation, information entropy, mean gradient, and spatial frequency of the fused images are quantitatively analyzed. The experimental results show that the false detection rate and missed detection rate of the model trained by the algorithm in this paper are significantly reduced compared with other algorithms, and the detection accuracy is improved from 75.51% to 88.86% compared with the existing algorithm using unimodal images, and the image processing speed can meet the demand of real-time detection.