Depth Estimation of Traffic Scenes from Image Sequence Using Deep Learning

被引：1

作者：

Liu, Xiaoxu ^{[1
]}

Yan, Wei Qi ^{[1
]}

机构：

[1] Auckland Univ Technol, Auckland 1010, New Zealand

来源：

IMAGE AND VIDEO TECHNOLOGY 2022, PSIVT 2022 | 2023年 / 13763卷

关键词：

Deep learning; automatic car; scene depth understanding; depth estimation; NEURAL-NETWORK;

D O I：

10.1007/978-3-031-26431-3_15

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Autonomous cars can accurately perceive the deployment of traffic scenes and the distance between visual objects in the scenarios through understanding the depth. Therefore, the depth estimation of scenes is a crucial step in the obstacle avoidance and pedestrian protection from autonomous vehicles. In this paper, a method for stereo depth estimation based on image sequences is introduced. In this project, we improve the performance of deep learning-based model by combining depth hints algorithm and MobileNetV2 encoder to enhance the loss function and increases computing speed. To the best of our knowledge, this is the first time MobileNetV2 is applied to depth estimation based on KITTI dataset.

引用

页码：186 / 196

页数：11