Traffic Flow Forecast based on Vehicle Count
Real-time traffic predictions have now become a time-being need for efficient traffic management due to the exponentially increasing traffic congestion. In this paper, a more pragmatic traffic management system is introduced to address traffic congestion, especially in countries such as Sri Lanka where there is no proper traffic monitoring database. Here the real-time traffic monitoring is performed using TFmini Plus light detection and ranging (LiDAR) sensor and vehicle count for next five minutes will be predicted by feeding consecutively collected data into the LSTM neural network. More than ten separate prediction models were trained, varying both window size and the volume of input data delivered to train the models. Since the accuracy results of all prediction models were above 70%, it demonstrates that this system can produce accurate predictions even if it is trained using less input data collection. Similarly the sensor accuracy test also resulted in 89.7% accuracy.
