# Gated2Depth: Real-time Dense Lidar from Gated Images

**Authors:** Tobias Gruber, Frank Julca-Aguilar, Mario Bijelic, Werner Ritter, Klaus Dietmayer, Felix Heide  
**arXiv:** [1902.04997](https://arxiv.org/abs/1902.04997)  
**Date:** 2019-02-13  
**Summary page:** https://www.summarizepaper.com/en/paper/1902.04997v1/

## Key points

- Authors: Tobias Gruber, Frank Julca-Aguilar, Mario Bijelic, Werner Ritter, Klaus Dietmayer, Felix Heide
- Introduction of innovative imaging framework "Gated2Depth" that transforms gated camera images into high-resolution depth maps
- Achieves depth resolution comparable to pulsed lidar measurements and overcomes limitations of existing scanning lidar systems
- Framework leverages learned architecture to extract dense depth information from three temporally gated images
- Exploits semantic context across gated slices within the architecture for effectiveness
- Trained on synthetic discriminator loss function without relying on dense depth labels for real-time performance
- Validated through extensive testing on outdoor driving scenarios and simulations
- Represents a significant advancement in imaging technology for generating dense depth maps in real-time without compromising accuracy or efficiency

## Summary

In their paper titled "Gated2Depth: Real-time Dense Lidar from Gated Images," authors Tobias Gruber, Frank Julca-Aguilar, Mario Bijelic, Werner Ritter, Klaus Dietmayer, and Felix Heide introduce an innovative imaging framework that transforms three images captured by a gated camera into high-resolution depth maps. This novel approach achieves depth resolution comparable to pulsed lidar measurements and addresses the limitations of existing scanning lidar systems. <br><br>

Traditional lidar technologies often suffer from high costs, power consumption, and bulky form factors associated with short-pulsed laser sources and scanning mechanics. In contrast, the proposed framework leverages a learned architecture to extract high-fidelity dense depth information from three temporally gated images acquired using a flash source and a high-resolution CMOS sensor. <br><br>

Central to the effectiveness of this approach is the exploitation of semantic context across gated slices within the architecture. By training on a synthetic discriminator loss function without relying on dense depth labels, the method achieves real-time performance. Essentially transforming a gated camera into a cost-effective alternative to traditional lidar systems,<br><br>

"Gated2Depth" is validated through extensive testing on various outdoor driving scenarios and simulations. Overall, it represents a significant advancement in imaging technology by offering a practical solution for generating dense depth maps in real-time without compromising on accuracy or efficiency. The research conducted by Gruber et al. opens up new possibilities for enhancing autonomous driving systems and other applications reliant on precise depth perception in complex environments.