Vulnerable Roadside User Protection System

Accurate, low footprint detection with vulnerable roadside users (VRUs) and road signs covered in the Automotive Tier 1 autonomous driving initiative.

Automotive Tier 1

2018

Client and Industry

A leader in the automotive industry based in Germany, renowned for its innovation, reliability, and quality. As a top global supplier, this company has significantly advanced automotive technology with its developments in fuel injection systems, safety features, and electric drives, reinforcing its essential role in the sector's progress.

Industry

Automotive Tier 1

Business Type

Enterprise

Services

Research & Development

Our Role

In a collaborative venture involving two leading automotive companies aimed at robotaxi development, we developed an advanced pedestrian protection system. This system leverages radar, ultrasonic, and video sensors to early identify Vulnerable Roadside Users (VRUs), assisting drivers and informing advanced automated driving systems.

Project Goals

01

Early Detection: Identify VRUs around the vehicle within the hazard range, alerting the driver or triggering automatic emergency braking if there is no timely response

02

Segmentation of VRUs: Classify VRUs into categories such as school guards, police officers, construction workers, and cyclists, enhancing context-aware Advanced Driver Assistance Systems (ADAS).

03

Intent Estimation: Estimate the intentions of VRUs through stable pose estimates, crucial for the development of future automated driving systems, such as predicting whether a person at a crosswalk will actually cross.

Business
Challenges

01.

Given the critical importance of VRUs on the road, the system must function flawlessly at all times. Achieving accuracy 99,999 times out of 100,000 is deemed insufficient.

02.

The VRU detector works alongside various systems like traffic sign and light detection, using a backbone network with specialized modules designed for multi-task learning. This setup requires coordination across 15 different teams to ensure updates to the model do not adversely affect their outputs. Challenges include retraining, versioning, and testing within this complex framework.

03.

In high-speed scenarios, vehicles must detect VRUs from afar to allow safe deceleration and stopping. Despite using high-resolution cameras, the challenge arises because objects at great distances appear significantly smaller (less than 50px). Balancing the demands of swift detection, significant distance, and camera constraints poses a complex engineering challenge.

Solution

We developed a VRU detection and protection system combining radar, ultrasonic sensors, and cameras to enable prompt braking or manoeuvring to avoid pedestrian collisions. Upon initiating evasive action, it activates steering support and alerts drivers within half a second, processing accident scenarios in just 5 milliseconds for maximum safety.

Key Features

01

Enhanced Radar Sensors: These sensors, with advanced signal processing, accurately detect objects like pedestrians and cyclists, even in limited visibility or adverse weather, by analysing their direction and speed relative to the vehicle.

02

Early Detection Capability: Enables the system to alert the driver or automatically engage emergency braking to prevent collisions or significantly lower impact speeds, thus reducing serious injury risks.

03

VRU Pose-SSD1: Incorporates cutting-edge models for VRU detection and 2-D pose estimation, tailored for robotaxis, providing accurate pose estimates and enabling sophisticated analyses of VRU behaviours to predict their intentions.

04

Robust Detection: Operates effectively in adverse weather and tracks multiple objects simultaneously, enhancing safety.

05

NCAP Standards Compliance: Meets standards for automatic emergency braking, effectively safeguarding vulnerable road users.

References

1 - C. Kumar, J. Ramesh, B. Chakraborty, R. Raman, C. Weinrich, A. Mundhada, A. Jain, F. Flohr , “VRU Pose-SSD: Multiperson Pose Estimation For Automated Driving” in proceedings of Innovative Applications of Artificial Intelligence (IAAI 2021), Virtual, February 2021. PDF

Impact

01.

Research Collaboration: Co-authored the research paper "VRU Pose-SSD: Multiperson Pose Estimation for Automated Driving" with Bosch, Mercedes, and the Indian Institute of Science.

02.

NCAP Compliance: The predictive pedestrian protection system meets NCAP requirements for automatic emergency braking.

03.

Potential of Automatic Emergency Braking: Shows significant potential in preventing or mitigating frontal collisions with pedestrians at speeds up to 60 km/h, significantly reducing injury risks, avoiding or mitigating half of the accidents with cyclists resulting in personal injury in Germany, and reducing up to 30% of relevant pedestrian accidents.