Pedestrian Detection Using Moving Camera

Abstract: In the paper, segmentation is performed using Sliding Window method. When the camera is moving, then in such cases, the prevalent methods for segmentation like background subtraction will get failed. Therefore, with the help of sliding window technique which scans through the frame through different scales, segmentation of the object of interest, which is the pedestrian in our case, is done. For this, due to the advantages offered by HOG such as better representation of human contour, invariance to illumination changes and small movements, and easy computation in constant time make it best suited for its application as a feature extractor. After feature extraction, Neural network classifier modelled on training and subsequent testing, decides whether the region cropped by sliding window is in actual containing a pedestrian or not. In addition to that, neural network is preferred because single neural classifier can be used for training and classification of multiple classes and also for large set of database, convergence is better in neural network based classifier. And from the results it is also proven that neural network is successful in pedestrian detection.

Keywords: ANN, Canny HOG, PDS, RD-HOG.