Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Directed-CP: Directed Collaborative Perception for Connected and Autonomous Vehicles via Proactive Attention

Yihang Tao, Senkang Hu, Zhengru Fang, Yuguang Fang

TL;DR

Directed-CP addresses uneven traffic distribution and limited communication budgets by enabling ego CAVs to proactively specify directions of interest and to adaptively fuse features from neighbors using RSU-informed masks and a direction-aware attention mechanism. The framework comprises RSU-aided direction masking, a direction control module with sparse query maps, and a direction-weighted detection loss to guide training toward targeted directions. Empirical results on the V2X-Sim 2.0 dataset show substantial gains in local directional perception (PD-IoU) for interested directions and modest improvements in overall AP under the same budget, validating the efficiency of directed CP under resource constraints. Overall, Directed-CP offers a practical approach to directed, budget-aware collaborative perception for connected autonomous vehicles, enabling more targeted and efficient information fusion in dynamic traffic scenarios.

Abstract

Collaborative perception (CP) leverages visual data from connected and autonomous vehicles (CAV) to enhance an ego vehicle's field of view (FoV). Despite recent progress, current CP methods expand the ego vehicle's 360-degree perceptual range almost equally, which faces two key challenges. Firstly, in areas with uneven traffic distribution, focusing on directions with little traffic offers limited benefits. Secondly, under limited communication budgets, allocating excessive bandwidth to less critical directions lowers the perception accuracy in more vital areas. To address these issues, we propose Direct-CP, a proactive and direction-aware CP system aiming at improving CP in specific directions. Our key idea is to enable an ego vehicle to proactively signal its interested directions and readjust its attention to enhance local directional CP performance. To achieve this, we first propose an RSU-aided direction masking mechanism that assists an ego vehicle in identifying vital directions. Additionally, we design a direction-aware selective attention module to wisely aggregate pertinent features based on ego vehicle's directional priorities, communication budget, and the positional data of CAVs. Moreover, we introduce a direction-weighted detection loss (DWLoss) to capture the divergence between directional CP outcomes and the ground truth, facilitating effective model training. Extensive experiments on the V2X-Sim 2.0 dataset demonstrate that our approach achieves 19.8\% higher local perception accuracy in interested directions and 2.5\% higher overall perception accuracy than the state-of-the-art methods in collaborative 3D object detection tasks. Codes are available at https://github.com/yihangtao/Directed-CP.git.

Directed-CP: Directed Collaborative Perception for Connected and Autonomous Vehicles via Proactive Attention

TL;DR

Directed-CP addresses uneven traffic distribution and limited communication budgets by enabling ego CAVs to proactively specify directions of interest and to adaptively fuse features from neighbors using RSU-informed masks and a direction-aware attention mechanism. The framework comprises RSU-aided direction masking, a direction control module with sparse query maps, and a direction-weighted detection loss to guide training toward targeted directions. Empirical results on the V2X-Sim 2.0 dataset show substantial gains in local directional perception (PD-IoU) for interested directions and modest improvements in overall AP under the same budget, validating the efficiency of directed CP under resource constraints. Overall, Directed-CP offers a practical approach to directed, budget-aware collaborative perception for connected autonomous vehicles, enabling more targeted and efficient information fusion in dynamic traffic scenarios.

Abstract

Collaborative perception (CP) leverages visual data from connected and autonomous vehicles (CAV) to enhance an ego vehicle's field of view (FoV). Despite recent progress, current CP methods expand the ego vehicle's 360-degree perceptual range almost equally, which faces two key challenges. Firstly, in areas with uneven traffic distribution, focusing on directions with little traffic offers limited benefits. Secondly, under limited communication budgets, allocating excessive bandwidth to less critical directions lowers the perception accuracy in more vital areas. To address these issues, we propose Direct-CP, a proactive and direction-aware CP system aiming at improving CP in specific directions. Our key idea is to enable an ego vehicle to proactively signal its interested directions and readjust its attention to enhance local directional CP performance. To achieve this, we first propose an RSU-aided direction masking mechanism that assists an ego vehicle in identifying vital directions. Additionally, we design a direction-aware selective attention module to wisely aggregate pertinent features based on ego vehicle's directional priorities, communication budget, and the positional data of CAVs. Moreover, we introduce a direction-weighted detection loss (DWLoss) to capture the divergence between directional CP outcomes and the ground truth, facilitating effective model training. Extensive experiments on the V2X-Sim 2.0 dataset demonstrate that our approach achieves 19.8\% higher local perception accuracy in interested directions and 2.5\% higher overall perception accuracy than the state-of-the-art methods in collaborative 3D object detection tasks. Codes are available at https://github.com/yihangtao/Directed-CP.git.
Paper Structure (13 sections, 8 equations, 5 figures, 3 tables)

This paper contains 13 sections, 8 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Collaborative perception
  4. Attention-based LiDAR perception
  5. Methodology
  6. RSU-aided direction masking
  7. Direction-aware selective attention
  8. Direction-weighted detection loss
  9. experiments
  10. Experimental setup
  11. Quantitative results
  12. Qualitative results
  13. conclusion

Figures (5)

  • Figure 1: Overview of directed CP framework. Under limited communication budget, ego CAV enhances perception in complex traffic directions while maintaining basic awareness in low-traffic areas.
  • Figure 2: Problem overview. Ego CAV combines RSU's DAS with its interests to create direction mask. The mask, along with initial query map, neighboring CAVs' poses, and communication budget, are input to QC-Net. QC-Net contains: (i) Direction Control Module, generating direction-prioritized query confidence maps (QCMs), and (ii) Query Clipping Layer, selecting top $Q_{max} \times H \times W$ queries based on QCMs under budget constraints.
  • Figure 3: AP of different CP methods under various communication budgets.
  • Figure 4: Visualization comparison on V2X-Sim 2.0 dataset. Green and red boxes denote ground truth and predictions, respectively. While Where2comm improves global perception over lower-bound, it shows limitations in certain directions. Directed-CP enhances perception in ego's interested directions (marked as 1, with right arrow showing ego CAV's movement).
  • Figure 5: Attention weight visualization on neighboring CAVs. Where2comm distributes attention equally for omnidirectional perception, while Directed-CP focuses on features relevant to ego's interested directions.