Detecting dangerous takeovers on a bicycle

In this repo the usage of a recurrent neural network for detecting when a car overtakes a bicycle at a dangerously close distance based on low resolution depth videos is explored.

Contents

This repo contains code for:

• ./training/: training and testing the neural network. This folder also contains the recorded and labeled trainingsdata.
• ./deployment/sensebox/: deploying the trained network on a senseBox MCU S2 with a VL53L8CX sensor for recording the depth images
• ./data-collection/: collecting more data for training the neural network

Concept

Hardware setup on a bicycle carrier