Vehicle Deteection And Classification Algorithm Based On TMR Sensor

In the field of intelligent transportation, vehicle detection and classification based on magnetic sensor is a hot issue. Tunnel Magneto Resistance(TMR) sensor, the forth generation magnetic sensor, has higher sensitivity and smaller size than Anisotropic Magneto Resistance(AMR) sensor. Accordingly, vehicle classification algorithm research based on TMR sensor is promising and necessary.This dissertation reports a vehicle detection and classification system by using TMR sensor. The sensor nodes are put aside by the road. Magnetic sensor node, based on Earth’s magnetic field variation detection, includes TMR sensor module and ZigBee wireless network module. Signal amplifier and potential shift module are designed to adjust the sensor’s output voltage. CC2530, as the MCU, collects the adjusted analog signal from TMR sensor and transmits the signals by ZigBee wireless network. In order to obtain magnetic variation signals when vehicle go through the nodes, Adaptive Threshold Detection Algorithm(ATDA) based on Baseline Tracing is used for vehicle detection. We determine weight coefficient and threshold for the detection algorithm by comparison test. The screened vehicle magnetic signal can be basis for vehicle classification algorithm. For vehicle classification, a method for how to form signal based feature vector is presented. In addition, training data set and test data set for classification algorithm are constructed from massive vehicle magnetic signal samples.To classify vehicle into different types, we first study Back Propagation Neural Network algorithm. Experiments are made to find appropriate neural number of each layer of the three-layer network. The network consists of input layer, hidden layer and output layer. Based on magnetic signal, we test the classification accuracy of network with different transfer function and training function. The best combination of function parameters is provided. Secondly, Directed Acyclic Graph algorithm is used to classify vehicle type. According to the difference among the typical signals, a module based on DAG algorithm is built. Kernel function is determined according to classification accuracy of experiment result.Abundant signal data is collected to test both classification algorithm. Algorithm with better classification accuracy is determined for the vehicle detection and classification system.