Object detection via background modeling is important in preprocessing step before many video-based automatic vehicle analysis systems. Because it contributes high computational cost portion. Furthermore, it also results in high memory usage and creates a constraint in the classification process when it used in video streaming application. This paper provides a contemporary approach to vehicle detection and classification for highway traffic surveillance video. In this works, we are creating a system to detect the vehicles and automatically classifying them into three desired classes as Car, SUV, and Truck. The proposed approach combines Incremental Principle Component Pursuit (IPCP) background subtraction for vehicle detection via background subtraction method and it complies with Ensemble subspace -Nearest Neighbor (EsKNN) classifier. A combination of these methods is used to acquire optimal result in the feature extraction process by using HOG corner detection. IPCP works by modeled background one frame at a time while it is capable to adapt to changes in the background. With this characteristic, it is suitable for real-time processing background subtraction system. Then, the outcomes are classified using ensemble-based classifiers technique, a predictive based model with employs KNN as building blocks. The proposed method is evaluated using three datasets of common highway surveillance video using testing datasets, which obtained from changedetect.net. Comparing with other direct detection and classification technique our method has to achieve an outstanding result and delivers the accuracy of 96.5%, the highest among the tested methods.