As a prerequisite for autonomous driving, scene understanding has attracted extensive research. With the rise of convolutional neural network (CNN)-based deep learning technique, research on achieved significant progress. This paper aims to provide comprehensive survey learning-based approaches in driving. We categorize these works into four work streams, including object detection, full semantic segmentation, instance and lane line segmentation. discuss analyze according their...

Lane detection is a challenging task due to problems like the diversity of lanes, occlusion, dazzle light, and so on. We believe that two factors are helpful solve above therefore improve performance, including global context dependency effective feature representation focusing on important channels. In this work, we propose an instance segmentation approach develop novel dual attention network DALane- Net for real-time lane detection. The leverages spatial channel mechanism achieve better...

Visual odometry (VO) refers to incremental estimation of the motion state an agent (e.g., vehicle and robot) by using image information, is a key component modern localization navigation systems. Addressing monocular VO problem, this paper presents novel end-to-end network for camera ego-motion. The learns latent subspace optical flow (OF) models sequential dynamics so that constrained relations between images. We compute OF field consecutive images extract representation in self-encoding...

<title>Abstract</title> Fruit recognition is critical for automated harvesting systems. This paper aims to explore a high-precision, real-time method strawberry recognition. Existing methods are mostly pixel-level and box-based. However, these cannot accurately or in determine the position of strawberries. Therefore, this proposes new contour-based detection segmentation network strawberries, called "StrawSnake." A strawberry-specific octagonal contour designed based on shape strawberries...

Motion estimation is crucial to predict where other traffic participants will be at a certain period of time, and accordingly plan the route ego-vehicle. This paper presents novel approach estimate motion state by using region-level instance segmentation extended Kalman filter (EKF). involves three stages object detection, tracking parameter estimate. We first use accurately locate region for latter two stages. The combines color, temporal (optical flow), spatial (depth) information as basis...

Abstract Quantitative measurement of smartphone screen scratches is crucial for pricing in the used market. Traditional manual visual inspection methods suffer from inherent limitations, namely being labor-intensive, subjective, and prone to inaccuracy. Hence, this study proposes a vision-based method as viable solution overcome these challenges. The algorithm uses Hessian enhancement extract scratch features, applies adaptive thresholding distinguish features background, employs...

This paper proposes a novel unsupervised learning framework for depth recovery and camera ego-motion estimation from monocular video. The exploits the optical flow (OF) property to jointly train models. Unlike existing methods, our method extracts features rather than raw RGB images, thereby enhancing learning. In addition, we exploit forward-backward consistency check of generate mask invalid region in image, accordingly, eliminate outlier regions such as occlusion moving objects...

This paper presents a novel unsupervised learning framework for estimating scene depth and camera pose from video sequences, fundamental to many high-level tasks such as 3D reconstruction, visual navigation, augmented reality. Although existing methods have achieved promising results, their performance suffers in challenging scenes those with dynamic objects occluded regions. As result, multiple mask technologies geometric consistency constraints are adopted this research mitigate negative...

Automated harvesting systems rely heavily on precise and real-time fruit recognition, which is essential for improving efficiency reducing labor costs. Strawberries, due to their delicate structure complex growing environments, present unique challenges automated recognition systems. Current methods predominantly utilize pixel-level box-based approaches, are insufficient applications inability accurately pinpoint strawberry locations. To address these limitations, this study proposes...

Addressing on monocular visual odometry problem, this paper presents a novel end-to-end network for estimation of camera ego-motion. The learns the latent space optical flow (OF) and models sequential dynamics so that motion is constrained by relations between images. We compute OF field consecutive images extract representation in self-encoding manner. A Recurrent Neural Network then followed to examine changes, i.e., conduct learning. extracted used regression 6-dimensional pose vector....