• convolutional neural network
• noise pollution
• quantization
• real-time
• traffic flow
• vehicle tracking

Traffic flow and speed are among the most important parameters needed to be estimated when studying noise pollution. In this thesis, a real-time multiple-vehicle recognition system was developed using a low-cost system consisting of a quantized convolutional network with a tracking algorithm, capable of counting the amount of vehicles passing through and their associated acceleration and speed information.
The results obtained through the proposed methodology provide a low-cost but yet powerful tool for the corresponding authorities involved in the task of noise pollution assessment.