Identification of Pedestrians From Confused Planar Objects Using Light Field Imaging

Recognizing a fake pedestrian on a planar surface (2-D fake pedestrian) is an essential and challenging task in the field of machine vision because of its wide range of applications. In this paper, a 2-D fake pedestrian recognition method is proposed based on light-field (LF) imaging and support vector machine (SVM). This method can recognize a 2-D fake pedestrian with only one sensor in a single exposure. To evaluate the method, we construct a new pedestrian dataset comprising more than 1000 samples using LF imaging. The experimental results show that the highest accuracy of this method is greater than 96%. Moreover, due to the efficient SVM classifier and obvious shape difference between a 2-D fake pedestrian and the real one in LF depth images, an accessible accuracy (85%) will be obtained even if the number of training samples is not large (120 training samples). This preliminary work demonstrates that the proposed approach is quite valid for the recognition of a 2-D fake pedestrian and has potential application in the machine vision field.