Vision-based Automated Parking

Note : As of the end of December 2021

Vision-based Automated Parking

Abstract

Abstract

Panasonic is developing 3D sensing technology for automated parking system. Accurate and dense map is generated by Panasonic "Dense SLAM*" and 1-stroke parking in the various parking lots will be achieved.

SLAM* : Simultaneous Localization and Mapping

Features

Features

1 stroke, wide detection range APS, APS in challenging use cases, Detect small objects and avoid collision

1 stroke, wide detection range APS*

Utilize surround view cameras.
No additional sensors are needed.
Detection of parking slots on both sides enables 1 stroke parking.

APS* in challenging use cases

It covers various use cases such as narrow, misaligned parking space and parking space surrounded by bushes etc.
94% success rate in 500 test-drives on 20 use cases including challenging scenarios. (benchmark: 48 % success rate with sonar-based APS*)

Detect small objects and avoid collision

Small static objects (Pole, chain) can be detected- Complete parking maneuver without any collision risks.

APS* : Automated Parking System

Technical Advantages

Technical Advantages

Dense SLAM for fisheye camera

Dense SLAM for fisheye camera

6,000 points per camera in 30 fps
(2,000 points per camera in conventional SFM*)

Precise map with High precision, High coverage

Spatial error rate < 5%
Coverage distance : ~15m
(depending on camera spatial resolution)

Embedded friendly

Processing time benchmark : 27 msec in TDA2x (70 msec for conventional SFM*)
Scalable / configurable framerate and point cloud density for various target hardware.

SFM* : Structure From Motion

Applications

Applications

Back-over Protection
Automated/Valet parking
Autonomous Driving
Vision-based Automated Parking  Applications