RESUMEN

High-precision maps are widely applied in intelligent-driving vehicles for localization and planning tasks. The vision sensor, especially monocular cameras, has become favoured in mapping approaches due to its high flexibility and low cost. However, monocular visual mapping suffers from great performance degradation in adversarial illumination environments such as on low-light roads or in underground spaces. To address this issue, in this paper, we first introduce an unsupervised learning approach to improve keypoint detection and description on monocular camera images. By emphasizing the consistency between feature points in the learning loss, visual features in dim environment can be better extracted. Second, to suppress the scale drift in monocular visual mapping, a robust loop-closure detection scheme is presented, which integrates both feature-point verification and multi-grained image similarity measurements. With experiments on public benchmarks, our keypoint detection approach is proven robust against varied illumination. With scenario tests including both underground and on-road driving, we demonstrate that our approach is able to reduce the scale drift in reconstructing the scene and achieve a mapping accuracy gain of up to 0.14 m in textureless or low-illumination environments.

RESUMEN

Accurate image feature point detection and matching are essential to computer vision tasks such as panoramic image stitching and 3D reconstruction. However, ordinary feature point approaches cannot be directly applied to fisheye images due to their large distortion, which makes the ordinary camera model unable to adapt. To address such a problem, this paper proposes a self-supervised learning method for feature point detection and matching on fisheye images. This method utilizes a Siamese network to automatically learn the correspondence of feature points across transformed image pairs to avoid high annotation costs. Due to the scarcity of the fisheye image dataset, a two-stage viewpoint transform pipeline is also adopted for image augmentation to increase the data variety. Furthermore, this method adopts both deformable convolution and contrastive learning loss to improve the feature extraction and description of distorted image regions. Compared with traditional feature point detectors and matchers, this method has been demonstrated with superior performance on fisheye images.

Asunto(s)

Algoritmos , Procesamiento de Imagen Asistido por Computador , Procesamiento de Imagen Asistido por Computador/métodos , Aprendizaje Automático Supervisado

RESUMEN

We propose a completely unsupervised approach to simultaneously estimate scene depth, ego-pose, ground segmentation and ground normal vector from only monocular RGB video sequences. In our approach, estimation for different scene structures can mutually benefit each other by the joint optimization. Specifically, we use the mutual information loss to pre-train the ground segmentation network and before adding the corresponding self-learning label obtained by a geometric method. By using the static nature of the ground and its normal vector, the scene depth and ego-motion can be efficiently learned by the self-supervised learning procedure. Extensive experimental results on both Cityscapes and KITTI benchmark demonstrate the significant improvement on the estimation accuracy for both scene depth and ego-pose by our approach. We also achieve an average error of about 3° for estimated ground normal vectors. By deploying our proposed geometric constraints, the IOUaccuracy of unsupervised ground segmentation is increased by 35% on the Cityscapes dataset.