LiDAR-Based Multi-Task Road Perception Network for Autonomous Vehicles

Yan, Fuwu; Wang, Kewei; Zou, Bin; Tang, Luqi; Li, Wenbo; Lv, Chen

doi:10.1109/access.2020.2993578

Cited by 20 publications

(11 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Lower MAE but higher RMSE means that our network has less average error than that of [4], but has more pixels with large error. To calculate the iRMSE and iMAE, we just have to reverse the y predict and y true in (1) and (2).…”

Section: Discussionmentioning

confidence: 99%

“…In addition, LiDAR performance does not depend on the changes in lighting conditions. These advantages make LiDAR an ideal 3D sensor for outdoor applications such as autonomous vehicles [2] [3]. One drawback of LiDAR sensor is its data sparsity: When mapping a Velodyne 64 line LiDAR HDL-64E point cloud to its corresponding high-resolution image obtained from a camera, only about 10% of the pixels have depth values.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DepthNet: Real-Time LiDAR Point Cloud Depth Completion for Autonomous Vehicles

et al. 2020

View full text Add to dashboard Cite

Autonomous vehicles rely heavily on sensors such as camera and LiDAR, which provide real-time information about their surroundings for the tasks of perception, planning and control. Typically a LiDAR can only provide sparse point cloud owing to a limited number of scanning lines. By employing depth completion, a dense depth map can be generated by assigning each camera pixel a corresponding depth value. However, the existing depth completion convolutional neural networks are very complex that requires high-end GPUs for processing, and thus they are not applicable to real-time autonomous driving. In this paper, a light-weight network is proposed for the task of LiDAR point cloud depth completion. With an astonishing 96.2% reduction in the number of parameters, it still achieves comparable performance (9.3% better in MAE but 3.9% worse in RMSE) to the state-of-the-art network. For real-time embedded platforms, depthwise separable technique is applied to both convolution and deconvolution operations and the number of parameters decreases further by a factor of 7.3, with only a small percentage increase in RMSE and MAE performance. Moreover, a system-on-chip architecture for depth completion is developed on a PYNQ-based FPGA platform that achieves realtime processing for HDL-64E LiDAR at the speed 11.1 frame per second.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

DepthNet: Real-Time LiDAR Point Cloud Depth Completion for Autonomous Vehicles

et al. 2020

View full text Add to dashboard Cite

show abstract

“…For developing an occlusion-free road segmentation system, Yan et al [28] proposed a LiDAR-based LMRoadNet network where the network is trained using a weighted loss function. Using a 1/4 scale feature map, they can perform road ground height prediction and road topology detection simultaneously with reduced complexity, and their fusion strategy can expand the field of view of the autonomous driving system.…”

Section: Related Workmentioning

confidence: 99%

Estimation of Road Boundary for Intelligent Vehicles Based on DeepLabV3+ Architecture

Das¹,

Fime

Siddique

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Road boundary estimation is an essential task for autonomous vehicles and intelligent driving assistants. It is considerably straightforward to attain the task when roads are marked properly with indicators. However, estimating road boundary reliably without prior knowledge of the road, such as road markings, is extremely difficult. This paper proposes a method to estimate road boundaries in different environments with deep learning-based semantic segmentation, and without any predefined road markings. The proposed method employed an encoder-decoder-based DeepLab architecture for segmentation with different types of backbone networks such as VGG16, VGG19, ResNet-50, and ResNet-101 while handling the class imbalance problem by weighing the loss contribution of the model's different outputs. The performance of the proposed method is verified using the 'ICCV09DATA' dataset. The method outperformed other existing methods and achieved the accuracy, precision, recall, f-measure of 0.9596±0.0097, 0.9453±0.0118, 0.9369±0.0149, and 0.9408±0.0135 respectively while using RestNet-101 as a backbone network and Dice Coefficient as a loss function. The detailed experimental analysis confirms the feasibility of the proposed method for road boundary estimation in different challenging environments.

show abstract

“…Here the authors suggest to use two sets of images relative to the intersection processed with DNN and RNN respectively. Regarding the LiDAR domain, the authors in [18] proposed a network called LMRoadNet aimed to simultaneously segment road surface and perform topology recognition by an aggregation of consecutive measures. Another interesting approach that exploits LiDAR was presented in [19].…”

Section: Related Workmentioning

confidence: 99%

“…1) KITTI: We used the work in [6] to select cam 02-03 color images, raw LiDAR readings and GPS-RTK positions of 8 residential sequences, six recorded on 2011/09/30 [18,20,27,28,33,34] and two recorded on 2011/10/03 [27,34]. Frames were automatically chosen from the whole sequence by gathering only those that are close up-to 20m from the intersection center.…”

Section: A Datasetmentioning

confidence: 99%

Model Guided Road Intersection Classification

Ballardini,

Hernández,

Sotelo

2021

Preprint

View full text Add to dashboard Cite

Understanding complex scenarios from in-vehicle cameras is essential for safely operating autonomous driving systems in densely populated areas. Among these, intersection areas are one of the most critical as they concentrate a considerable number of traffic accidents and fatalities. Detecting and understanding the scene configuration of these usually crowded areas is then of extreme importance for both autonomous vehicles and modern Advanced Driver Assistance Systems (ADAS) aimed at preventing road crashes and increasing the safety of vulnerable road users. This work investigates intersection classification from RGB images using well-consolidate neural network approaches along with a method to enhance the results based on the teacher/student training paradigm. An extensive experimental activity aimed at identifying the best input configuration and evaluating different network parameters on both the well-known KITTI dataset and the new KITTI-360 sequences shows that our method outperforms current state-of-the-art approaches on a per-frame basis and prove the effectiveness of the proposed learning scheme.

show abstract

LiDAR-Based Multi-Task Road Perception Network for Autonomous Vehicles

Cited by 20 publications

References 33 publications

DepthNet: Real-Time LiDAR Point Cloud Depth Completion for Autonomous Vehicles

DepthNet: Real-Time LiDAR Point Cloud Depth Completion for Autonomous Vehicles

Estimation of Road Boundary for Intelligent Vehicles Based on DeepLabV3+ Architecture

Model Guided Road Intersection Classification

Contact Info

Product

Resources

About