Vehicle Detection using Dimensionality Reduction based on Deep Belief Network for Intelligent Transportation System

Vehicle detection is one of essential technology on Intelligent Transportation System. By detect the vehicle, we may know the presence and the number of vehicles on the road for some intervals of time. The challenge is the high dimensionality of features to represent a vehicle. To detect the vehicle, there are two processes, feature extraction and classification. The high dimensionality feature, which is produced on feature extraction step, make a heavy computation on the classification step. This research proposed a dimensionality reduction using Deep Belief Network (DBN) for vehicle detection. We try to detect cars and motorcycles. The feature extraction method is based on descriptive methods such as scaled invariant feature transform. DBN is used to reduce the high dimensionality features that is produced by descriptive methods. For classification task, we choose Support Vector Machine method. Moreover, UIUC dataset and our original data are chose to evaluate the proposed method performance. We are also compared DBN with Principal Component Analysis (PCA) and other method. The result indicates that using DBN as dimensionality reduction method performed better than PCA and others method in vehicle detection.