Raspberry Pi Applications

**301**

**Chapter 14**

*Toby White*

**Abstract**

accuracy score of over 90%.

tially also enhance the toy industry.

neural networks

**1. Introduction**

Computational Efficiency: Can

Pi Complete the Computations

Required to Follow the Path?

Something as Small as a Raspberry

This chapter explains the development processes of a prototype autonomous toy car. It focuses on the design and implementation of transforming a normal remote control toy car into a self-contained vehicle with the capability to drive autonomously. This would be proven by making it follow a track of any layout. It uses a neural network (NN) in the form of a multilayer perceptron (MLP) to process images in real time to generate a movement instruction. Upon completion, the vehicle demonstrated the ability to be able to follow a track of any layout, while staying between both sides of the track. The collision avoidance system proved to be effective up to a distance of 50 cm in front of the vehicle in order to let it stop prior to hitting an object. The neural network processing of the image in order to classify it in a real time proved to be above the expectation of around 5 FPS and has an

**Keywords:** raspberry pi, image recognition, classification, machine learning,

Given the recent development of self-driving cars by companies such as Tesla, Google and others, it was of interest to attempt to replicate this on a smaller scale, by implementing a similar method on a small electronic toy car. An issue many developers in this industry are having is the issue of leaving the device connected to a wide area network (WAN) all the time, leaving the vehicles vulnerable to not only hacking but also vulnerable to being unable to make decisions in out-of-reach places such as the countryside where a WAN may not be available or only available intermittently. The decision was made to attempt the replication on a small scale, using a closed network with the computations occurring within the vehicle, as opposed to externally. The study was viewed as a proof of concept to test the possibility and feasibility, which could help lead research on full-scale self-driving cars and poten-

This study mainly focuses on a proof-of-concept implementation of artificial

neural networks (ANNs) to classify road sign images in a real-time scenario
