Traffic scene awareness for intelligent vehicles using ConvNets and stereo vision

Guindel, Carlos; Martín, David; Armingol, José María

doi:10.1016/j.robot.2018.11.010

Cited by 25 publications

(21 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Apart from the customary horizontal flipping, scale jittering (from −10% to 15% of the original scale) has been applied to training images to improve the robustness of the resulting model. Additionally, the repeat factor sampling proposed in [ 35 ] is used to mitigate the effects of class imbalance. The resulting deep neural network model is aimed to provide 2D bounding box detections of the objects.…”

Section: Data Mining Process and Dataset Organizationmentioning

confidence: 99%

“… ‘Dominant orientation’ is obtained in several steps. Firstly, the object’s yaw angle is derived using both the viewpoint estimate given by the neural network and the location computed through the pinhole camera model, as in [ 35 ]. Afterward, following the approach employed for viewpoint estimation, the 360° range of possible yaw angles is discretized into eight bins, and the bin representing the orientation of the closest object becomes the value of the “dominant orientation” parameter.…”

Section: Data Mining Process and Dataset Organizationmentioning

confidence: 99%

See 1 more Smart Citation

Dataset Construction from Naturalistic Driving in Roundabouts

Cuenca

Guindel

Aliane

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

A proper driver characterization in complex environments using computational techniques depends on the richness and variety of data obtained from naturalistic driving. The present article proposes the construction of a dataset from naturalistic driving specific to maneuvers in roundabouts and makes it open and available to the scientific community for performing their own studies. The dataset is a combination of data gathered from on-board instrumentation and data obtained from the post-processing of maps as well as recorded videos. The approach proposed in this paper consists of handling roundabouts as a stretch of road that includes 100 m before the entrance, the internal part, and 100 m after the exit. This stretch of road is then spatially sampled in small sections to which data are associated.

show abstract

Section: Data Mining Process and Dataset Organizationmentioning

confidence: 99%

Section: Data Mining Process and Dataset Organizationmentioning

confidence: 99%

Dataset Construction from Naturalistic Driving in Roundabouts

Cuenca

Guindel

Aliane

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…Applying the operation defined by (4) to each channel of the left color image, a total of six intensity edge maps can be obtained. The intensity edge maps are sent into three sets of downsample blocks to obtain a pyramid of intensity edge features.…”

Section: B Depth Discontinuity Aware Super-resolution Subnetworkmentioning

confidence: 99%

“…Depth estimated from stereo images has been the core information for vision-based practical applications, such as obstacle avoidance for robot navigation [1], 3D scene reconstruction for augmented and virtual reality system [2], and 3D visual object tracking and location [3], [4]. Given a pair of pre-rectified stereo images, the target of stereo matching is to accurately compute a disparity value for each pixel in the reference image.…”

Section: Introductionmentioning

confidence: 99%

Learning Efficient Stereo Matching Network With Depth Discontinuity Aware Super-Resolution

2019

View full text Add to dashboard Cite

Deep convolutional neural networks (CNNs) have shown great potential to provide accurate depth estimation based on stereo images. Previous work has focused on developing robust stereo matching architectures, while little attention has been paid on improving the network efficiency. In this paper, we propose an efficient Siamese CNN architecture that combines the low resolution disparity estimation and the depth discontinuity aware super-resolution. Specifically, we propose to construct, filter and perform regression on a low resolution cost volume through the designed stereo matching backbone network. A fast depth discontinuity aware super-resolution subnetwork is proposed for upsampling the low resolution disparity map to the desired resolution. Under the guidance of the intensity edge features extracted from the left color image, depth edge residuals are hierarchically learned to refine the upsampled depth map. A delayed upsampling structure is designed to ensure that the computational complexity is proportional to the spatial size of the input disparity map. We also propose to supervise the first derivative loss of the predicted disparity map that makes the network adaptively aware of the depth discontinuity edges. Experiments show that the proposed stereo matching network achieves a comparable prediction accuracy and much faster running speed compared with state-of-the-art methods. INDEX TERMS Stereo matching network, disparity estimation, depth map super-resolution, depth discontinuity aware loss.

show abstract

“…The extended version of this paper, published in the Robotics and Autonomous Systems journal [114], is partially included in the thesis, as stated before.…”

Section: A C K N O W L E D G M E N T S / a G R A D E C I M I E N T O Smentioning

confidence: 99%