Nowadays, technologies of autonomous driving are very high-developed, but it is still not enough for fully safety usage on the public roads, almost in the cities. In this case it is very important to have more additional active and passive safety systems. And one of this systems is “car hearing”. An autonomous vehicle must have at least similar characteristics as a human driver, including perceptional skills like vision for object recognition and human control capabilities for driving. Most of the drivers can also hear, and it is a very important source of information for understanding the environment around the car.
First of all it is very important to recognize and localize different kinds of emergency sounds, such as ambulance or fireworks siren, hard breaking and human‘s shout in emergency cases. Second is passive dangers, such as overtaking car or motorcycle and car which is coming from behind.
In this topic we are going to develop a sound classification neural network which must be able to recognize different types of traffic sounds the data from 8 microphones installed on the top of the car. For example it is very important to recognize ambulance siren, human‘s shout or hard braking sound in emergency cases.
Master Thesis: Machine Learning for Sound Classification for autonomous driving
- The design, development and implementation of sound classification algorithms.
- Strong programming skills (Python), ROS
- Sound filtering and separation for sound processing
- TOA-based sound localization algorithms
- Neural networks types for sound processing
- Volume-based sound localization algorithms