Generic object detection framework with spatially pooled features

Abstract

Generic Object detection technique plays an important role in the traffic surveillance and security-related issues. Research has been done over the past several years and accomplished great progresses via convolutional neural network (CNN) which has greatly enhanced the performance in image classification and object detection. This proposal is similar to the notion of R-CNN [1], presents a novel method that combines the spatially pooled features (sp-Cov) as a part of aggregated channel (ACF) and CNN for object detection. The proposed technique uses sp-Cova and ACF to select the possible object on interest regions and then trains a CNN model to filter out non-face candidates. Merging the results of sp-Cov + ACF and CNN to get the final detection window(s). The proposed framework achieves the good performance with state-of-the-art methods on numerous benchmark data sets.