Speed Bot

Bug Fixes

A bug which caused the app to crash before launching was fixed! If you encounter any other bugs or issues, feel free to contact the developer at [email protected].

Speed Bot’s Aims

The end goal of Speed Bot is to reduce traffic casualties globally on a large scale. 1.35 million people die every year as a result of road traffic injuries. Two significant causes of this are distracted driving and speeding. 53% of drivers exceed the speed limit on 30mph roads and 81% on 20mph roads. Similarly, a quarter of drivers admit to using their phones whilst driving, causing 70 thousand casualties every year.

People are also too reliant on the data provided by GPS. Roads are constantly getting repaired, yet GPS satellites can’t receive this information immediately, whereas traffic signs can. In low-visibility scenarios, such as in fog or heavy foliage, traffic signs can easily be dismissed by the human eye, risking the safety of drivers, passengers, and pedestrians.

Introduction

Speed Bot is an app that detects and recognises traffic signs in your live camera feed. This is done using machine learning in Python and Kivy. However, the networks are currently nowhere near accurate and efficient enough to be used on the road. Similarly, only 56 different signs can be detected, a tiny fraction of the hundreds of signs found worldwide. Users need the 100% accuracy guarantee before they should rely on this app on the road, otherwise it has the possibility of causing more harm than relief.

An issue with neural networks is data. Unlike humans, who virtually understand objects as soon as they are seen, machines require hundreds of thousands of images to achieve the same result. Currently, there are only major traffic sign datasets from a handful of countries, as well as some smaller datasets elsewhere. This is not enough to ensure global safety. Thus, I propose a Crowdsourced Traffic Sign Dataset.

User Submission Program

The app will automatically save the 30 most confidently recognised signs and 10 pictures taken containing signs. Users are then given the option to upload these images and their labels to a server, which then adds these to their respective databases, ready to be trained upon. After this, the files are automatically deleted from users' devices. If the signs are labelled incorrectly, the app allows users to delete these images. This program is voluntary as it requires access to the Internet, yet we need as many people as possible joining the program. More training data = More accurate neural networks = Improved road safety for everyone.

How to use the app

On the default main screen, there should be two buttons: "Submit Recognition Data" and "Submit Detection Data". These buttons will direct users to their respective User Submission Programme screens. The app will automatically find signs when it’s in the main screen, and will locally save training data at specific intervals (you can change these intervals using the sliders in the settings).

In the top left corner of the screen, there is an icon that should open the menu. In the top right corner, there is a home icon that will take users back to the main screen.

Safety Precautions

Due to safety, users will be unable to turn sound on.

Requirements

Necessary Requirements: Camera, Storage

Optional: GPS for speedometer, Internet for User Submission Programs

Known Bugs

Chinese works largely perfectly, apart from one instance (when you click on any button in the settings page). There seems to also be an issue with the app crashing on load-up for some phones, so please do keep this in mind.