As a critical task in autonomous driving perception systems, 3D object detection is used to identify and track key objects, such as vehicles and pedestrians. However, detecting distant, small, or occluded objects (hard instances) remains a challenge, which directly compromises the safety of autonomous driving systems. We observe that existing multi-modal 3D object detection methods often follow a single-guided paradigm, failing to account for the differences in information density of hard instances between modalities. In this work, we propose DGFusion, based on the Dual-guided paradigm, which fully inherits the advantages of the Point-guide-Image paradigm and integrates the Image-guide-Point paradigm to address the limitations of the single paradigms. The core of DGFusion, the Difficulty-aware Instance Pair Matcher (DIPM), performs instance-level feature matching based on difficulty to generate easy and hard instance pairs, while the Dual-guided Modules exploit the advantages of both pair types to enable effective multi-modal feature fusion. Experimental results demonstrate that our DGFusion outperforms the baseline methods, with respective improvements of +1.0% mAP, +0.8% NDS, and +1.3% average recall on nuScenes. Extensive experiments demonstrate consistent robustness gains for hard instance detection across ego-distance, size, visibility, and small-scale training scenarios.

The sparse cross-modality detector offers more advantages than its counterpart, the Bird’s-Eye-View (BEV) detector, particularly in terms of adaptability for downstream tasks and computational cost savings. However, existing sparse detectors overlook the quality of token representation, leaving it with a sub-optimal foreground quality and limited performance. In this paper, we identify that the geometric structure preserved and the class distribution are the key to improving the performance of the sparse detector, and propose a Sparse Selector (SS). The core module of SS is Ray-Aware Supervision (RAS), which preserves rich geometric information during the training stage, and Class-Balanced Supervision, which adaptively reweights the salience of class semantics, ensuring that tokens associated with small objects are retained during token sampling. Thereby, outperforming other sparse multi-modal detectors in the representation of tokens. Additionally, we design Ray Positional Encoding (Ray PE) to address the distribution differences between the LiDAR modality and the image. Finally, we integrate the aforementioned module into an end-to-end sparse multi-modality detector, dubbed CrossRay3D. Experiments show that, on the challenging nuScenes benchmark, CrossRay3D achieves state-of-the-art performance with 72.4% mAP and 74.7% NDS, while running$1.84\times $faster than other leading methods. Moreover, CrossRay3D demonstrates strong robustness even in scenarios where LiDAR or camera data are partially or entirely missing. The code is available onhttps://github.com/xuehaipiaoxiang/CrossRay3D

Cooperative perception presents significant potential for enhancing the sensing capabilities of individual vehicles, however, inter-agent latency remains a critical challenge. Latencies cause misalignments in both spatial and semantic features, complicating the fusion of real-time observations from the ego vehicle with delayed data from others. To address these issues, we propose TraF-Align, a novel framework that learns the flow path of features by predicting the feature-level trajectory of objects from past observations up to the ego vehicle’s current time. By generating temporally ordered sampling points along these paths, TraF-Align directs attention from the current-time query to relevant historical features along each trajectory, supporting the reconstruction of current-time features and promoting semantic interaction across multiple frames. This approach corrects spatial misalignment and ensures semantic consistency across agents, effectively compensating for motion and achieving coherent feature fusion. Experiments on two real-world datasets, V2V4Real and DAIR-V2X-Seq, show that TraF-Align sets a new benchmark for asynchronous cooperative perception. The code is available at https://github.com/zhyingS/TraF-Align.

End-to-end autonomous driving frameworks enable seamless integration of perception and planning but often rely on one-shot trajectory prediction, which may lead to unstable control and vulnerability to occlusions in single-frame perception. To address this, we propose the Momentum-Aware Driving (MomAD) framework, which introduces trajectory momentum and perception momentum to stabilize and refine trajectory predictions. MomAD comprises two core components: (1) Topological Trajectory Matching (TTM) employs Hausdorff Distance to select the optimal planning query that aligns with prior paths to ensure coherence; (2) Momentum Planning Interactor (MPI) cross-attends the selected planning query with historical queries to expand static and dynamic perception files. This enriched query, in turn, helps regenerate long-horizon trajectory and reduce collision risks. To mitigate noise arising from dynamic environments and detection errors, we introduce robust instance denoising during training, enabling the planning model to focus on critical signals and improve its robustness. We also propose a novel Trajectory Prediction Consistency (TPC) metric to quantitatively assess planning stability. Experiments on the nuScenes dataset demonstrate that MomAD achieves superior long-term consistency (≥ 3s) compared to SOTA methods. Moreover, evaluations on the curated Turning-nuScenes shows that MomAD reduces the collision rate by 26% and improves TPC by 0.97m (33.45%) over a 6s prediction horizon, while closed- loop on Bench2Drive demonstrates an up to 16.3% improvement in success rate. The source code is available at https://github.com/adept-thu/MomAD.

With the rapid advancement of autonomous driving, safety concerns have become the primary barrier to its commercialization. Compliance with traffic laws is crucial for ensuring road safety. However, the current laws, formulated for human drivers, present challenges for autonomous systems due to ambiguous language description, complicating accurate judgment and government monitoring. It is imperative to transform traffic laws into machine-interpretable logical frameworks while simul-taneously resolving ambiguities in legal terminology to ensure clarity and precision. This study focuses on urban intersections, characterized by high traffic complexity and diverse participants. We propose a formalization method for right-of-way laws and develop a threshold analysis framework based on processed data from SIND, which rigorously defines the prioritization of right-of-way. The optimal compliance threshold is determined through sensitivity analysis, evaluated using the proposed Weighted TPN score (WTPNs). Meanwhile, the threshold was implemented in online monitoring at intersections. The dataset is available online via: https://github.comlSOTIF-AVLab/SinD

Modern autonomous vehicle perception systems often struggle with occlusions and limited perception range. Previous studies have demonstrated the effectiveness of cooperative perception in extending the perception range and overcoming occlusions, thereby enhancing the safety of autonomous driving. In recent years, a series of cooperative perception datasets have emerged; however, these datasets primarily focus on cameras and LiDAR, neglecting 4D Radar—a sensor used in single-vehicle autonomous driving to provide robust perception in adverse weather conditions. In this paper, to bridge the gap created by the absence of 4D Radar datasets in cooperative perception, we present V2X-Radar, the first large-scale, real-world multi-modal dataset featuring 4D Radar. V2X-Radar dataset is collected using a connected vehicle platform and an intelligent roadside unit equipped with 4D Radar, LiDAR, and multi-view cameras. The collected data encompasses sunny and rainy weather conditions, spanning daytime, dusk, and nighttime, as well as various typical challenging scenarios. The dataset consists of 20K LiDAR frames, 40K camera images, and 20K 4D Radar data, including 350K annotated boxes across five categories. To support various research domains, we have established V2X-Radar-C for cooperative perception, V2X-Radar-I for roadside perception, and V2X-Radar-V for single-vehicle perception. Furthermore, we provide comprehensive benchmarks across these three sub-datasets.

While most recent autonomous driving system focuses on developing perception methods on ego-vehicle sensors, people tend to overlook an alternative approach to leverage intelligent roadside cameras to extend the perception ability beyond the visual range. We discover that the state-of-the-art vision-centric detection methods perform poorly on roadside cameras. This is because these methods mainly focus on recovering the depth regarding the camera center, where the depth difference between the car and the ground quickly shrinks while the distance increases. In this paper, we propose a simple yet effective approach, dubbed BEVHeight++, to address this issue. In essence, we regress the height to the ground to achieve a distance-agnostic formulation to ease the optimization process of camera-only perception methods. By incorporating both height and depth encoding techniques, we achieve a more accurate and robust projection from 2D to BEV spaces. On popular 3D detection benchmarks of roadside cameras, our method surpasses all previous vision-centric methods by a significant margin. In terms of the ego-vehicle scenario, BEVHeight++ surpasses depth-only methods with increases of +2.8% NDS and +1.7% mAP on the nuScenes test set, and even higher gains of +9.3% NDS and +8.8% mAP on the nuScenes-C benchmark with object-level distortion. Consistent and substantial performance improvements are achieved across the KITTI, KITTI-360, and Waymo datasets as well.

Local Feature Matching, a pivotal component of numerous computer vision tasks (e.g., structure from motion and visual localization), has been effectively addressed by Transformer-based methods. Nevertheless, these methods solely incorporate long-range context information among keypoints with a fixed receptive field, which constrains the network from appropriately reconciling the importance of features with diverse receptive fields to realize complete image perception, hence limiting feature matching accuracy. In addition, these methods employ a conventional handcrafted encoding approach to incorporate positional information of keypoints into visual descriptors, which limits the capability of networks to extract effective positional encoding message. In this study, we propose FMRT, a novel detector-free method that reconciles local features with diverse receptive fields adaptively and utilizes parallel networks to realize reliable positional encoding. Specifically, FMRT proposes a dedicated reconciliatory transformer (RecFormer) that contains a global perception attention layer to identify visual descriptors with different receptive fields and integrate global context information under various scales, a perception weight layer to measure the importance of various receptive fields adaptively, and a local perception feed-forward network to extract deep aggregated multi-scale local feature representation. Moreover, we introduce a novel axis-wise position encoder (AWPE) that views positional encoding as two keypoints encoding tasks along the row and column dimensions, decouples the x- and y-coordinates of keypoints into two independent 1D vectors, and designs two parallel network branches to explicitly encodes geometric correlations among keypoints, hence realizing reliable positional encoding. Extensive experiments indicate that FMRT yields impressive performance on multiple tasks, including relative pose estimation, visual localization, homography estimation, and image matching. Besides, we integrate FMRT into a localization framework and conduct a visual localization experiment in a real scene, which further demonstrate the superiority of FMRT. Note to Practitioners—This paper presents a novel approach to enhancing the performance of local feature matching in computer vision tasks. Traditional methods often rely on fixed receptive fields for integrating context among keypoints, which can limit the perception of the complete image and, consequently, the precision of feature matching. Our work introduces a Reconciliatory Transformer that not only addresses these limitations by effectively reconciling the importance of features across varying receptive fields but also improves the integration of positional information into visual descriptors. The techniques developed here can be adapted to a wide range of systems, e.g., image matching for computer vision and visual localization for autonomous driving, offering practitioners a tool to significantly improve the fidelity of feature matching, which is foundational for accurate interaction with the surrounding environment.

Roadside perception can significantly enhance the safety of autonomous vehicles by extending their perceptual capabilities beyond the visual range and addressing occluded regions. However, current state-of-the-art vision-based roadside detection methods exhibit high accuracy on labeled scenes but perform poorly on new scenes. This limitation arises because roadside cameras remain stationary after installation and can only gather data from a single scene, leading the algorithm to overfit these roadside backgrounds and camera positions. To tackle this issue, we propose an innovativeScenarioGeneralization Framework forVision-based Roadside3DObject Detection, calledSGV3D. Specifically, we utilize a Background-suppressed Module (BSM) to reduce background overfitting in vision-centric pipelines by diminishing background features during the 2D to bird’s-eye-view projection. Furthermore, by introducing the Semi-supervised Data Generation Pipeline (SSDG) that employs unlabeled images from new scenes, we generate diverse foreground instances with varying camera poses, mitigating the risk of overfitting to specific camera positions. Experiments conducted on two large-scale roadside benchmarks demonstrate that SGV3D, with only a minimal increase in latency, effectively improves the scenario generalization capabilities of vision-based roadside 3D object detectors. The code is available here (https://github.com/yanglei18/SGV3D).

LiDAR and camera are complementary sensors for 3D object detection in autonomous driving. However, it is challenging to explore the unnatural interaction between point clouds and images, and the critical factor is how to conduct feature alignment of these heterogeneous modalities. Currently, many methods achieve feature alignment through projection calibration, without accounting for the impact of sensors misalignment errors, resulting in sub-optimal performance. In this paper, we present GraphAlign++, a more accurate feature alignment framework for 3D object detection by graph matching. Specifically, we construct the nearest neighbor relationship by calculating Euclidean distances of point cloud features within the subspaces. Through the projection calibration between the image and point cloud pairs, we project the nearest neighbors of point cloud features onto the corresponding image. Then by matching the nearest neighbors of a single point-feature of the point cloud with multiple pixel-features of the image, we search for a more appropriate feature alignment. In addition, we provide a self-attention module to enhance the weights of significant relations to fine-tune the feature alignment between these two heterogeneous modalities. Extensive experiments on nuScenes benchmark demonstrate the effectiveness and efficiency of GraphAlign++. Notably, due to the more accurate feature alignment, which contributes to increase mAP by 3.10% on KITTI test hard level, our method is remarkably beneficial for long-range object detection.

LiDAR-based sparse 3-D object detection plays a crucial role in autonomous driving applications due to its computational efficiency advantages. Existing methods either use the features of a single central voxel as an object proxy or treat an aggregated cluster of foreground points as an object proxy. However, the former cannot aggregate contextual information, resulting in insufficient information expression in object proxies. The latter relies on multistage pipelines and auxiliary tasks, which reduce the inference speed. To maintain the efficiency of the sparse framework while fully aggregating contextual information, in this work, we propose SparseDet that designs sparse queries as object proxies. It introduces two key modules: the local multiscale feature aggregation (LMFA) module and the global feature aggregation (GFA) module, aiming to fully capture the contextual information, thereby enhancing the ability of the proxies to represent objects. The LMFA module achieves feature fusion across different scales for sparse key voxels via coordinate transformations and using nearest neighbor relationships to capture object-level details and local contextual information, whereas the GFA module uses self-attention mechanisms to selectively aggregate the features of the key voxels across the entire scene for capturing scene-level contextual information. Experiments on nuScenes and KITTI demonstrate the effectiveness of our method. Specifically, SparseDet surpasses the previous best sparse detector VoxelNeXt (a typical method using voxels as object proxies) by 2.2% mean average precision (mAP) with 13.5 frames/s on nuScenes and outperforms VoxelNeXt by 1.12%$\text {AP}_{\text {3-D}}$on hard level tasks with 17.9 frames/s on KITTI. What is more, not only the mAP of SparseDet exceeds that of FSDV2 (a classical method using clusters of foreground points as object proxies) but also its inference speed is 1.3 times faster than FSDV2 on the nuScenes test set. The code has been released inhttps://github.com/liulin813/SparseDet.git.

In the realm of modern autonomous driving, the perception system is indispensable for accurately assessing the state of the surrounding environment, thereby enabling informed prediction and planning. The key step to this system is related to 3D object detection that utilizes vehicle-mounted sensors such as LiDAR and cameras to identify the size, the category, and the location of nearby objects. Despite the surge in 3D object detection methods aimed at enhancing detection precision and efficiency, there is a gap in the literature that systematically examines their resilience against environmental variations, noise, and weather changes. This study emphasizes the importance of robustness, alongside accuracy and latency, in evaluating perception systems under practical scenarios. Our work presents an extensive survey of camera-only, LiDAR-only, and multi-modal 3D object detection algorithms, thoroughly evaluating their trade-off between accuracy, latency, and robustness, particularly on datasets like KITTI-C and nuScenes-C to ensure fair comparisons. Among these, multi-modal 3D detection approaches exhibit superior robustness, and a novel taxonomy is introduced to reorganize the literature for enhanced clarity. This survey aims to offer a more practical perspective on the current capabilities and the constraints of 3D object detection algorithms in real-world applications, thus steering future research towards robustness-centric advancements.

Although the majority of recent autonomous driving systems concentrate on developing perception methods based on ego-vehicle sensors, there is an overlooked alternative approach that involves leveraging intelligent roadside cameras to help extend the ego-vehicle perception ability beyond the visual range. We discover that most existing monocular 3D object detectors rely on the ego-vehicle prior assumption that the optical axis of the camera is parallel to the ground. However, the roadside camera is installed on a pole with a pitched angle, which makes the existing methods not optimal for roadside scenes. In this paper, we introduce a novel framework for Roadside Monocular 3D object detection with ground-aware embeddings, named MonoGAE. Specifically, the ground plane is a stable and strong prior knowledge due to the fixed installation of cameras in roadside scenarios. In order to reduce the domain gap between the ground geometry information and high-dimensional image features, we employ a supervised training paradigm with a ground plane to predict high-dimensional ground-aware embeddings. These embeddings are subsequently integrated with image features through cross-attention mechanisms. Furthermore, to improve the detector’s robustness to the divergences in cameras’ installation poses, we replace the ground plane depth map with a novel pixel-level refined ground plane equation map. Our approach demonstrates a substantial performance advantage over all previous monocular 3D object detectors on widely recognized 3D detection benchmarks for roadside cameras. The code and pre-trained models will be released soon.

Multiple object tracking (MOT) is becoming increasingly significant for autonomous driving and intelligent transportation systems. However, traditional MOT methods cannot track the objects accurately and robustly due to the lack of effective feature extraction and data association in complex traffic scenarios. In this paper, we propose a novel joint detection and tracking method AttentionTrack by introducing multiple features attention. Firstly, we design a self-motivated feature extraction attention network (FEAN) to adaptively produce effective decoupled features for detection and tracking tasks in different scenarios. Secondly, we build a spatial-temporal data association (STDA) framework to achieve more accurate and robust tracking by considering the historical features of trajectory through different times. Moreover, we conduct comprehensive experiments on the KITTI, UA-DETRAC and MOT17 benchmarks, and the results show that our approach achieves competitive performance compared with the state-of-the-art (SOTA) trackers.

Road surface conditions, especially geometry profiles, enormously affect driving performance of autonomous vehicles. Vision-based online road reconstruction promisingly captures road information in advance. Existing solutions like monocular depth estimation and stereo matching suffer from modest performance. The recent technique of Bird’s-Eye-View (BEV) perception provides immense potential to more reliable and accurate reconstruction. This paper uniformly proposes two simple yet effective models for road elevation reconstruction in BEV named RoadBEV-mono and RoadBEV-stereo, which estimate road elevation with monocular and stereo images, respectively. The former directly fits elevation values based on voxel features queried from image view, while the latter efficiently recognizes road elevation patterns based on BEV volume representing correlation between left and right voxel features. Insightful analyses reveal their consistence and difference with the perspective view. Experiments on real-world dataset verify the models’ effectiveness and superiority. Elevation errors of RoadBEV-mono and RoadBEV-stereo achieve 1.83 cm and 0.50 cm, respectively. Our models are promising for practical road preview, providing essential information for promoting safety and comfort of autonomous vehicles. The code is released athttps://github.com/ztsrxh/RoadBEV.

While most recent autonomous driving system focuses on developing perception methods on ego-vehicle sensors, people tend to overlook an alternative approach to leverage intelligent roadside cameras to extend the perception ability beyond the visual range. We discover that the state-of-the-art vision-centric bird's eye view detection methods have inferior performances on roadside cameras. This is because these methods mainly focus on recovering the depth regarding the camera center, where the depth difference between the car and the ground quickly shrinks while the distance increases. In this paper, we propose a simple yet effective approach, dubbed BEVHeight, to address this issue. In essence, instead of predicting the pixel-wise depth, we regress the height to the ground to achieve a distance-agnostic formulation to ease the optimization process of camera-only perception methods. On popular 3D detection benchmarks of roadside cameras, our method surpasses all previous vision-centric methods by a significant margin. The code is available at https://github.com/ADLab-AutoDrive/BEVHeight.

LiDAR and cameras are complementary sensors for 3D object detection in autonomous driving. However, it is challenging to explore the unnatural interaction between point clouds and images, and the critical factor is how to conduct feature alignment of heterogeneous modalities. Currently, many methods achieve feature alignment by projection calibration only, without considering the problem of coordinate conversion accuracy errors between sensors, leading to sub-optimal performance. In this paper, we present GraphAlign, a more accurate feature alignment strategy for 3D object detection by graph matching. Specifically, we fuse image features from a semantic segmentation encoder in the image branch and point cloud features from a 3D Sparse CNN in the LiDAR branch. To save computation, we construct the nearest neighbor relationship by calculating Euclidean distance within the subspaces that are divided into the point cloud features. Through the projection calibration between the image and point cloud, we project the nearest neighbors of point cloud features onto the image features. Then by matching the nearest neighbors with a single point cloud to multiple images, we search for a more appropriate feature alignment. In addition, we provide a self-attention module to enhance the weights of significant relations to fine-tune the feature alignment between heterogeneous modalities. Extensive experiments on nuScenes benchmark demonstrate the effectiveness and efficiency of our GraphAlign.

Semi-supervised learning (SSL) has promising potential for improving model performance using both labelled and unlabelled data. Since recovering 3D information from 2D images is an ill-posed problem, the current state-of-the-art methods of monocular 3D object detection (Mono3D) have relatively low precision and recall, making semi-supervised learning for Mono3D tasks challenging and understudied. In this work, we propose a unified and effective semi-supervised learning framework called Mix-Teaching that can be applied to most monocular 3D object detectors. Based on the idea of decomposition and recombination, unlabelled samples are firstly decomposed into collections of image patches with high-quality predictions and collections of background images containing no objects. The student model is then trained on the mixed images containing dense instances with high-quality pseudo-labels generated by the recombination operation. In addition, we propose an uncertainty-based filter to distinguish high-quality pseudo-labels from noisy predictions during the decomposition process. As results in KITTI and nuScenes benchmarks, Mix-Teaching consistently improves MonoFlex and GUPNet by significant margins under various labeling ratios. Our method achieves around +6.34%$AP_{3D}$improvement against the GUPNet on the validation set when using only 10% labelled data. Using the full training set and the additional 38K raw images from KITTI, it can further improve the MonoFlex by +4.65% absolute improvement on$AP_{3D}$for car detection, reaching 18.54%$AP_{3D}$, which ranks the 1st place among all monocular based methods on the KITTI test leaderboard.

3D Multi-object tracking (MOT) ensures consistency during continuous dynamic detection, conducive to subsequent motion planning and navigation tasks in autonomous driving. However, camera-based methods suffer in the case of occlusions and it can be challenging to track the irregular motion of objects for LiDAR-based methods accurately. Some fusion methods work well but do not consider the untrustworthy issue of appearance features under occlusion. At the same time, the false detection problem also significantly affects tracking. As such, we propose a novel camera-LiDAR fusion 3D MOT framework based on Combined Appearance-Motion Optimization (CAMO-MOT), which uses both camera and LiDAR data and significantly reduces tracking failures caused by occlusion and false detection. For occlusion problems, we are the first to propose an occlusion head to select the best object appearance features multiple times effectively, reducing the influence of occlusions. To decrease the impact of false detection in tracking, we design a motion cost matrix based on confidence scores which improve the positioning and object prediction accuracy in 3D space. As existing multi-object tracking methods always evaluate each category separately and do not consider the mismatch between objects of different categories, we also propose to build a multi-category cost to implement multi-object tracking in multi-category scenes. A series of validation experiments are conducted on the KITTI and nuScenes tracking benchmarks. Our proposed method achieves state-of-the-art performance with 79.99% HOTA and the lowest identity switches (IDS) value (23 for Car and 137 for Pedestrian) among all multi-modal MOT methods on the KITTI test dataset. And our method achieves state-of-the-art performance among all algorithms on the nuScenes test dataset with 75.3% AMOTA.