Acoustic detection of rear approaching vehicles for cyclists

Thesis Abstract

The project aims to detect rear approaching vehicles for cyclist with a low power consumption. Study focuses on acoustic features of the sound of rear approaching vehicles and examines the useful indicators to detect the vehicles.

The project includes more then one correlation and reveals their success rates for as many as samples possible. The project is given as a proposal for increasing the safety of cyclists riding on the country roads.

It can be sometimes hazardous for bike riders to ride either on the same road with the cars or on the suburban road.Thus, an alerting system which will be mounted on the bicycle is required.

The aim of the project is to find most suitable sensor which can generate the alert early enough that cyclist can take an action before the vehicle reaches up.Since the sensor will be used for bicycles, power consumption and the size of the sensor should be rather small.

Audio detection was chosen from among all the possible detection types.In order to distinguish the sound from an approaching vehicle among all environmental noises,many samples were collected and analyzed.Classifying and comparing the sounds; the most interesting features of the vehicle approaches were found out as; frequency shifts and amplitude increment in power spectrum, regularity/irregularity changes within the sound, mass point shifts and loudness level increment.

Four different methods were used regarding correlations.Samples were filtered to obtain better results.Considering all the cases, there is a threshold was set for all methods in order not to confuse them with environmental noises or risk less cases.Specifying a threshold value has also its disadvantages,because making alert generation more reliable is inversely proportional with success ratio.The most of the methods given in this report have the approximate success ratio of 70% for each method is quite sufficient, considering those methods can be fused together to have more accurate sensor.

My work is a proof that many different types of sound correlations can be used to determine if there is a vehicle approaching or not.Methods that pointed out in this report can be used to build up a fine alerting system to increases the awareness of cyclists.