PI-RCNN: An Efficient Multi-Sensor 3D Object Detector with Point-Based Attentive Cont-Conv Fusion Module

Xie, Liang; Xiang, Chao; Yu, Zi‐Tao; Xu, Guodong; Zheng, Yang; Cai, Deng; He, Xiaofei

doi:10.1609/aaai.v34i07.6933

Cited by 159 publications

(94 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, the fusion methods based on BEV or voxel format are not accurate enough. Thus, PI-RCNN [74] proposes a novel fusion method named Point-based Attentive Cont-conv Fusion module to fuse multisensor features directly on 3D points. Except for continuous convolution, Point-Pooling and Attentive Aggregation are used to fuse features expressively.…”

Section: Multi-sensor Fusion-based 3d Object Detection Methodsmentioning

confidence: 99%

A Review of 3D Object Detection for Autonomous Driving of Electric Vehicles

Dai

Chen

Bao

et al. 2021

WEVJ

View full text Add to dashboard Cite

In recent years, electric vehicles have achieved rapid development. Intelligence is one of the important trends to promote the development of electric vehicles. As a result, autonomous driving system is becoming one of the core systems of electric vehicles. Considering that environmental perception is the basis of intelligent planning and safe decision-making for intelligent vehicles, this paper presents a survey of the existing perceptual methods in vehicles, especially 3D object detection, which guarantees the reliability and safety of vehicles. In this review, we first introduce the role of perceptual module in autonomous driving system and a relationship with other modules. Then, we classify and analyze the corresponding perception methods based on the different sensors. Finally, we compare the performance of the surveyed works on public datasets and discuss the possible future research interests.

show abstract

Section: Multi-sensor Fusion-based 3d Object Detection Methodsmentioning

confidence: 99%

A Review of 3D Object Detection for Autonomous Driving of Electric Vehicles

Dai

Chen

Bao

et al. 2021

WEVJ

View full text Add to dashboard Cite

show abstract

“…Instead, some methods [10], [13], [36] perform the point-wise detection on the point-wise features provided by the PointNet++ [16] or graph neural networks [37]. The point-wise point cloud features can also be augmented with the camera image features by projecting the points onto the image plane [38], [39]. To improve the orientation coverage of the cubic anchor, a novel spherical anchor for point cloud space is proposed in STD [11], but the box recall is still lagging behind that of the mapview RPN in the voxel-based methods [14], [15].…”

Section: D Object Detection Based On Pointsmentioning

confidence: 99%

P2V-RCNN: Point to Voxel Feature Learning for 3D Object Detection From Point Clouds

Sun

Luo

et al. 2021

IEEE Access

View full text Add to dashboard Cite

The most recent 3D object detectors for point clouds rely on the coarse voxel-based representation rather than the accurate point-based representation due to a higher box recall in the voxelbased Region Proposal Network (RPN). However, the detection performance is severely restricted by the information loss of pose details in the voxels and the variability in the relationship between the visible part and the full view of objects because of the perspective issue in data acquisition. In this paper, we propose a point-to-voxel feature learning approach to voxelize the point cloud with high-level point-wise semantic features, which enables the voxel-wise and the point-wise feature learning. We propose an attentive corner aggregation module to attentively aggregate the local point cloud surrounding a 3D proposal from the perspectives of eight corners of the proposal 3D bounding box. The experimental results on the competitive KITTI 3D object detection benchmark show that the proposed method achieves state-of-the-art performance.INDEX TERMS 3D object detection, point clouds, attention mechanism, autonomous driving.

show abstract

“…The LiDAR points in a frustum proposal are used to generate the instance segmentation and 3D bounding boxes by PointNet++ [14]. PI-RCNN [22] uses a point-based attentive contfuse module to fuse features from multiple sensors. Pseudo-LiDAR++ [23] uses stereo camera images with LiDAR points to generate dense pseudo point clouds and enhance the performance of 3D detectors.…”

Section: Related Workmentioning

confidence: 99%

“…Our method is at the same level of performance as the two-stage detector PI-RCNN [22] and the one-stage anchor-based detector SCNet [39]. We visualize the detect results in Figure 4.…”

Section: Experiments On the Kitti Validation Setmentioning

confidence: 99%

One-Stage Anchor-Free 3D Vehicle Detection from LiDAR Sensors

Zhao

et al. 2021

Sensors

View full text Add to dashboard Cite

Recent one-stage 3D detection methods generate anchor boxes with various sizes and orientations in the ground plane, then determine whether these anchor boxes contain any region of interest and adjust the edges of them for accurate object bounding boxes. The anchor-based algorithm calculates the classification and regression label for each anchor box during the training process, which is inefficient and complicated. We propose a one-stage, anchor-free 3D vehicle detection algorithm based on LiDAR point clouds. The object position is encoded as a set of keypoints in the bird’s-eye view (BEV) of point clouds. We apply the voxel/pillar feature extractor and convolutional blocks to map an unstructured point cloud to a single-channel 2D heatmap. The vehicle’s Z-axis position, dimension, and orientation angle are regressed as additional attributes of the keypoints. Our method combines SmoothL1 loss and IoU (Intersection over Union) loss, and we apply (cosθ,sinθ) as angle regression labels, which achieve high average orientation similarity (AOS) without any direction classification tricks. During the target assignment and bounding box decoding process, our framework completely avoids any calculations related to anchor boxes. Our framework is end-to-end training and stands at the same performance level as the other one-stage anchor-based detectors.

show abstract

PI-RCNN: An Efficient Multi-Sensor 3D Object Detector with Point-Based Attentive Cont-Conv Fusion Module

Cited by 159 publications

References 21 publications

A Review of 3D Object Detection for Autonomous Driving of Electric Vehicles

A Review of 3D Object Detection for Autonomous Driving of Electric Vehicles

P2V-RCNN: Point to Voxel Feature Learning for 3D Object Detection From Point Clouds

One-Stage Anchor-Free 3D Vehicle Detection from LiDAR Sensors

Contact Info

Product

Resources

About