NeurAda: Combining artificial neural network and Adaboost for accurate object detection

Object detection is a very important technique in computer vision which is mainly used in many applications. Many papers have addressed this problem and proposed different methods to improve the accuracy of detectors. The main disadvantages of common methods in object detection are high time complexity, wrong object detection, not detecting objects. Extracting features and classification are two step of detecting objects. In this paper, a new method is presented to improve some of the disadvantages using Histograms of Oriented Gradient (HOG) as feature extractor and artificial neural network combined with Adaboost (NeurAda) as a classifier to cover weak points of previous works. To evaluate the proposed method, NeurAda was compared to the three top obtained results of Pascal VOC 2011 methods in three categories. NeurAda improved car detection by 8.6%, bicycle detection by 0.8% and pedestrian detection by 5.2% in comparison to best results of Pascal VOC 2011.