基于可见光与红外热图像的行车环境复杂场景分割

复杂场景分割是自动驾驶领域智能感知的重要任务,对稳定性和高效性都有较高的要求.由于一般的场景分割方法主要针对可见光图像,分割效果非常依赖于图像获取时的光线与气候条件,且大多数方法只关注分割性能,忽略了计算资源.本文提出一种基于可见光与红外热图像的轻量级双模分割网络(DMSNet),通过提取并融合两种模态图像的特征得到最终分割结果.考虑到不同模态特征空间存在较大差异,直接融合将降低对特征的利用率,本文提出了双路特征空间自适应(DPFSA)模块,该模块能够自动学习特征间的差异从而转换特征至同一空间.实验结果表明,本文方法提高了对不同模态图像的利用率,对光照变化有更强的鲁棒性,且以少量参数取得了较好的分割性能.

Complex Scene Segmentation Based on Visible and Thermal Images in Driving Environment

Complex scene segmentation is an important task of intelligent perception in the field of autonomous driving,which has high requirements for stability and efficiency.Since general scene segmentation methods mainly focus on visible images,the segmentation result is highly dependent on the light and weather conditions at the time of image acquisition,and most methods only focus on segmentation performance and ignore computing resources.This paper proposes a lightweight dual model segmentation network(DMSNet)based on visible and thermal images,which can extract and fuse the features of the two modal images to obtain a final segmentation result.For large differences in the feature spaces of different modalities,direct fusion will reduce the utilization of features.This paper proposes a dual-path feature space adaptation(DPFSA)module,which can automatically learn the differences among features and convert them to the same space.The experimental results show that this method can better utilize the inherent information between different modal images.Moreover,the proposed method is more robust to illumination changes and can achieve good segmentation performance with only a small number of parameters.