在无人车夜视红外视频彩色化问题中, 考虑到可同时利用单帧图像的信息和视频的帧间信息, 提出了一种双通道循环生成对抗网络（DcCCAN）对夜视红外视频进行彩色化。DcCCAN是在循环一致生成对抗网络（CCAN）的基础上提出的双通道生成网络。双通道生成网络具有良好的图像特征提取能力, 能够自动提取视频中待处理图像的特征, 同时提取先前模型所生成图像的特征, 然后将特征信息整合后生成一幅目标图像。通过在生成对抗性训练中引入循环一致性训练机制, 可无监督地学习得到红外域图像到彩色域图像的映射关系, 从而实现红外视频的彩色化。实验表明该方法能够为视频中的红外图像赋予自然的色彩信息和纹理信息, 且满足实时性要求。

In the task of unmanned vehicle infrared video colorization, considering the uniqueness of a single frame and the continuity of the entire infrared video, a dual-channel cycle-consistent adversarial network (DcCCAN) based colorization method is proposed. The dual-channel generation network we proposed is on the basis of cycle-consistent adversarial network (CCAN) and has good image feature extraction ability, which can automatically extract the features of the frame in the video, and at the same time can extract the features of the previous frame generated. By joint training of adversarial loss and cycle-consistent loss, the function from infrared domain image to color domain image can be learned by unsupervised learning methods and the colorization of the infrared video can be realized. The experimental results show that the proposed method can provide natural color information and texture information for the infrared images in the video, and meet the real-time requirements.