Audio Scene Analysis For Transportation Monitoring And Urban Safety
- French Team Leader: Pierre Beauseroy, Laboratory of Systems Modeling and Dependability, UTT
- Chinese Team Leader: Zhu MengYao, School of Communication and Information Technology, SHU
- Team: Pierre Beauseroy, Zhu MengYao
In urban monitoring, the audio scene analysis is not an effectively complement to video scene reco- gnition but is a promising substitute. The main technology is the magnetic loop but it remains quite expensive because it requires civil works and maintenance and the accuracy is not very reliable also. An alternative technology is the measure of traffic with video camera but this technology is also expensive due, at the same time, to cost of cameras and the integration requirement in the infrastructure. Moreover cameras require a large bandwidth to send data to the processing center and the performances of the processing are dependant of the weather and lighting conditions. Cameras are also more intrusive in private life which makes the acceptability in dense urban areas less obvious.
In contrast, audio sensors are relatively cheap, less invasive for citizen privacy and we could imagine that requirements of integration in the infrastructure will be less restrictive. The amount of data to transmit to the processing center will be necessarily less important. The performances of the processing will not be affected at least by the lighting conditions.
In this project, we focus on audio scene analysis in the field of road traffic estimation and road traffic event detection and localization. The first aim of this project is to demonstrate the feasibility of an approach of (1) traffic flow estimation (speed and density) and (2) traffic event detection (huge noise may cause by accident or big truck) using audio signal. To achieve this goal, we will: Build small audio acquisition network, collection of traffic sound data and set up a database ; Analyze the feature of sound signal, try to find the extraction pattern of vehicle sound ; Discriminate situations, machine learning methods are used to classify the relevant situations (traffic jam, no traffic, abnormal situation like accident, etc.).