# Polarimetric Imaging for Perception

**Authors:** Michael Baltaxe, Tomer Pe'er, Dan Levi  
**arXiv:** [2305.14787](https://arxiv.org/abs/2305.14787)  
**Date:** 2023-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.14787v1/

## Key points

- Study by Michael Baltaxe, Tomer Pe'er, and Dan Levi on "Polarimetric Imaging for Perception"
- Emphasis on utilizing RGB-polarimetric camera for autonomous driving and driver-assistance systems
- Comparison of RGB-polarimetric camera to RGB camera in perception tasks
- Introduction of new dataset with RGB-polarimetric images, lidar scans, GNSS/IMU readings, and free space segmentations
- Focus on consistent readings of light polarization regardless of vehicle orientation
- Importance of free space detection and depth estimation for creating surround views
- Potential benefits of polarimetric imaging in improving perception algorithms for autonomous driving systems

## Summary

In their study titled "Polarimetric Imaging for Perception," Michael Baltaxe, Tomer Pe'er, and Dan Levi explore the potential benefits of utilizing an RGB-polarimetric camera in perception tasks related to autonomous driving and advanced driver-assistance systems. The researchers highlight the underutilization of light polarization as a valuable property for enhancing perception tasks. Specifically focusing on monocular depth estimation and free space detection during midday when light polarization is independent of the vehicle's heading, they compare the performance of an RGB-polarimetric camera to that of an RGB camera using state-of-the-art deep neural networks with minimal architectural changes. Additionally, they introduce a new dataset comprising RGB-polarimetric images, lidar scans, GNSS/IMU readings, and free space segmentations to support the development of perception algorithms leveraging light polarization. Acknowledging the complexity of light polarization influenced by various factors such as light properties, scene properties, and viewing direction, the study focuses on consistent readings regardless of the vehicle's orientation. Free space detection plays a crucial role in identifying drivable areas within an image while depth estimation provides a detailed distance map from the camera sensor to objects in each pixel. These functionalities are essential for creating surround views of the environment. Overall, Baltaxe et al. 's research underscores the potential advantages of incorporating polarimetric imaging into perception algorithms for autonomous driving systems and highlights its relevance in improving key functionalities such as free space detection and depth estimation. Their findings contribute valuable insights towards advancing technology for safer and more efficient automated driving experiences.