Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Camera-Only Bird's Eye View Perception: A Neural Approach to LiDAR-Free Environmental Mapping for Autonomous Vehicles

Anupkumar Bochare

TL;DR

This work presents a camera-only BEV perception system that extends the Lift-Splat-Shoot framework to fuse multi-view camera data with DepthAnythingV2 monocular depth and YOLOv11 object detection, producing a unified BEV semantic map and object placements without LiDAR. A depth-aware lifting, precise 3D projection using calibrated extrinsics, and BEV feature aggregation enable 360-degree scene understanding, while BEVLoss jointly optimizes segmentation, detection, and depth accuracy. Evaluated on OpenLane-V2 and NuScenes, the method achieves about 85% road segmentation accuracy and 85-90% vehicle detection relative to LiDAR ground truth, with average positional errors around 1.2 m, highlighting substantial cost savings (≈85% reduction) and practical viability for LiDAR-free autonomous navigation. The approach demonstrates strong gains from integrating DepthAnythingV2 and YOLOv11 within a unified BEV pipeline, with notable performance near LiDAR-based systems and clear directions for improving robustness, range, and efficiency in real-world deployments.

Abstract

Autonomous vehicle perception systems have traditionally relied on costly LiDAR sensors to generate precise environmental representations. In this paper, we propose a camera-only perception framework that produces Bird's Eye View (BEV) maps by extending the Lift-Splat-Shoot architecture. Our method combines YOLOv11-based object detection with DepthAnythingV2 monocular depth estimation across multi-camera inputs to achieve comprehensive 360-degree scene understanding. We evaluate our approach on the OpenLane-V2 and NuScenes datasets, achieving up to 85% road segmentation accuracy and 85-90% vehicle detection rates when compared against LiDAR ground truth, with average positional errors limited to 1.2 meters. These results highlight the potential of deep learning to extract rich spatial information using only camera inputs, enabling cost-efficient autonomous navigation without sacrificing accuracy.

Camera-Only Bird's Eye View Perception: A Neural Approach to LiDAR-Free Environmental Mapping for Autonomous Vehicles

TL;DR

This work presents a camera-only BEV perception system that extends the Lift-Splat-Shoot framework to fuse multi-view camera data with DepthAnythingV2 monocular depth and YOLOv11 object detection, producing a unified BEV semantic map and object placements without LiDAR. A depth-aware lifting, precise 3D projection using calibrated extrinsics, and BEV feature aggregation enable 360-degree scene understanding, while BEVLoss jointly optimizes segmentation, detection, and depth accuracy. Evaluated on OpenLane-V2 and NuScenes, the method achieves about 85% road segmentation accuracy and 85-90% vehicle detection relative to LiDAR ground truth, with average positional errors around 1.2 m, highlighting substantial cost savings (≈85% reduction) and practical viability for LiDAR-free autonomous navigation. The approach demonstrates strong gains from integrating DepthAnythingV2 and YOLOv11 within a unified BEV pipeline, with notable performance near LiDAR-based systems and clear directions for improving robustness, range, and efficiency in real-world deployments.

Abstract

Autonomous vehicle perception systems have traditionally relied on costly LiDAR sensors to generate precise environmental representations. In this paper, we propose a camera-only perception framework that produces Bird's Eye View (BEV) maps by extending the Lift-Splat-Shoot architecture. Our method combines YOLOv11-based object detection with DepthAnythingV2 monocular depth estimation across multi-camera inputs to achieve comprehensive 360-degree scene understanding. We evaluate our approach on the OpenLane-V2 and NuScenes datasets, achieving up to 85% road segmentation accuracy and 85-90% vehicle detection rates when compared against LiDAR ground truth, with average positional errors limited to 1.2 meters. These results highlight the potential of deep learning to extract rich spatial information using only camera inputs, enabling cost-efficient autonomous navigation without sacrificing accuracy.
Paper Structure (32 sections, 7 equations, 5 figures, 4 tables)

This paper contains 32 sections, 7 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Bird's Eye View Representations
  4. Camera-Only Perception Approaches
  5. Monocular and Multi-View Depth Estimation
  6. Object Detection for Autonomous Vehicles
  7. Methodology
  8. System Overview
  9. Multi-Camera Setup
  10. Lift-Splat-Shoot Implementation
  11. Depth-Aware Feature Extraction
  12. 3D Space Projection
  13. BEV Feature Aggregation
  14. Object Detection Integration
  15. Depth Estimation with DepthAnythingV2
  16. ...and 17 more sections

Figures (5)

  • Figure 1: System architecture diagram showing the complete pipeline from multi-camera inputs to BEV output with depth estimation and feature extraction components.
  • Figure 2: Visualization of BEV outputs for different methods, showing our approach compared to LiDAR-based and other camera-based methods.
  • Figure 3: The figure demonstrates our method's ability to accurately detect road boundaries, lane markings, and objects in the Bird's Eye View representation.
  • Figure 4: Examples of our system's performance in challenging scenarios, including night-time, rain, and complex urban environments.
  • Figure 5: Examples of failure cases, showing scenarios where our system underperforms compared to LiDAR-based approaches.