Object Tracking and Classification Using Millimeter-Wave Radar Based on LSTM

All-weather sensors are necessary to realize automated driving level 3 or more, one of which is a millimeter-wave radar. However, the radar has some issues such as low space resolution and noisy signal. In order to solve them, stochastic processing is necessary such as deep learning techniques. In this research, classification and tracking of target objects were tried by applying LSTM (Long Short Term Memory) which can treat time-series data. Reflection signals from the 76GHz radar for randomly moving objects which are cars, bicycles and pedestrians in the parking lot were measured, and they were tracked and classified by the classifier applying LSTM. Three types of input feature amount for the four and three-class classifiers with two types of LSTM were evaluated and compared. Then, the best algorithm and combination achieved high accuracy of 98.67% as the results. Furthermore, the classifier was evaluated by the dataset measured on the actual public road, and then no misclassification was occurred for crossing pedestrians at the signalized intersection. This indicates high generalization ability of this classifier.