Written by 6:40 pm Meetup

WSDC 5 – car accident risk prediction, RC, top-down view from side cameras


Warsaw Self-Driving Cars the 5th Meetup! Only one presentation is available online.

The 3rd presentation during WSDC 5

Here is the link to the event:


Thursday, May 30, 2019, 6:00 PM

Grochowska 306/308 Warsaw, PL

54 Members Went

WSDC 5 English slides + PL or EN presentations related to self-driving. 18:00 – Google Street View image of a house predicts car accident risk of its resident; Kinga Kita-Wojciechowska 18:40 – RC to Full Autonomy – The Journey; Paweł Misiewicz 19:20 – Modelling the top-down view from side cameras in the CARLA simulator; Maciej Dziubiński 20:00 Pizz…

Check out this Meetup →

WSDC 5 – Presentations

The first presentation was:
Google Street View image of a house predicts car accident risk of its resident
by Kinga Kita-Wojciechowska.

Kinga is a researcher at the University of Warsaw looking for innovation in insurance pricing through the application of AI. Prior to that, she was working 10 years in motor insurance pricing, most of the time at AXA Group, where she held various managerial and expert roles in Poland, France, Spain, Korea, Japan, and China. Holds a double master degree in mathematics and economics from the University of Warsaw.

Road traffic injuries are a leading cause of death worldwide. Proper estimation of car accident risk is critical for appropriate allocation of resources in healthcare, insurance, civil engineering, and other industries. We show how images of houses are predictive of car accidents. We analyze 20,000 addresses of insurance company clients, collect a corresponding house image using Google Street View, and annotate house features such as age, type, and condition. We find that this information substantially improves car accident risk prediction compared to the state-of-the-art risk model of the insurance company and could be used for price discrimination. From this perspective, public availability of house images raises legal and social concerns, as they can be a proxy of ethnicity, religion and other sensitive data.

Street view of houses in Poland
Examples of extremely different houses located in the same zip-code and residents of which have the same expected claim frequency by the current insurer’s model. (https://arxiv.org/abs/1904.05270)

Link to the paper: https://arxiv.org/abs/1904.05270

The second presentation by Paweł Misiewicz:
RC to Full Autonomy – The Journey

Paweł Misiewicz is an A.I. and Autonomous Vehicles enthusiast who spends his free time on R&D in those fields. He obtained his master’s degree from Faculty of Mathematics, Informatics, and Mechanics of the University of Warsaw in the field of Distributed Systems. For almost 20 years he has been taking part in various projects for the largest Telcos and Banks in Poland as a technical lead. For the last 10 years, he has been specializing in Big Data and Data Warehousing projects.

Paweł: In this talk, I will summarize my journey of developing Fully Autonomous RC Car capable of driving in small, well-mapped areas – the project that I started almost one year ago. The work is by no means finished, but I hope to share some useful experience with everyone who wants to move from simulators to the real world. I will present project goals, my findings and challenges encountered during the preparation of the car, compute platform, data acquisition, and processing.

The last presentation and only one which was recorded is:
Modelling the top-down view from side cameras in the CARLA simulator
by Maciej Dziubiński

Maciek works as a machine learning engineer at Nethone – a company specializing in detecting fraudulent credit card transactions. But after work, he devotes his time to autonomous vehicles and is primarily interested in applying machine learning in that field.

Maciek: I will go through the details of building a model for predicting the top-down view from a camera hanging 100m above a car, based on input from four side cameras mounted on the car. The presentation will expand on what was already described in the blog post:
https://medium.com/asap-report/from-semantic-segmentation-to-semantic-birds-eye-view-in-the-carla-simulator-1e636741af3f but I’ll focus more on the topic „what didn’t work and why?”