Micro motion detection system for traffic target classification

The main way of traffic monitoring is through cameras, but it is easily affected by external lighting. The ground is subject to various natural and human movements, which generate small vibration signals. Using micro motion signals to perceive the environment is a new detection technology. This research design consists of micro motion acquisition hardware, data processing and time-frequency domain feature extraction algorithms, and machine learning based object classification algorithms to achieve scene perception of people and vehicles in smart transportation, and can also be used for security detection. After collecting field data, the 2935 effective slices extracted by the algorithm were trained and tested. The average accuracies of the MLP, SVM, GBT, and RF models were 95.808%, 91.039%, 95.570%, and 95.468%, respectively. The MLP (Multi-Layer Perceptron) with shallow neural network structure is the optimal model, with the highest average recognition rate and the smallest standard deviation. The experimental results show that this system is not affected by environmental factors such as weather, light intensity, and electromagnetic fields. It is compact and easy to deploy, with strong concealment, small data volume, and high reliability.