Multiscale information fusion-based deep learning framework for campus vehicle detection

Abstract

ABSTRACT Vehicle detection is a hotspot in the field of remote sensing image analysis. In particular, campus vehicle detection can assess the density of traffic in an area and provide security for students. The detection accuracy is low for dense vehicle areas or complex background areas. According to the feature of campus vehicle, we propose a multiscale information fusion strategy to construct a novel deep learning framework for campus vehicle detection. This new method based on Single Shot MultiBox Detector (SSD) combines a lightweight deep neural network MobileNet to extract features. A sub-network composed of multiple convolutional layers is connected to detect and locate the object. This method fuses feature information on multiple levels. When removing overlapped object candidate regions, the threshold value is set based on the non-maximum suppression method to eliminate redundant candidate regions. Therefore, the generated negative samples are reduced, which guarantees the stable effect of the proposed model. Experiments show that the proposed vehicle detection method has a faster detection speed. The robustness and accuracy of the proposed model are better than other related vehicle detection methods.