Search

There are two main contributions in this article. One is that an extraction means of dominant frequencies is proposed for the first time. The footstep events (FEs) from diverse subjects are analyzed comparatively in frequency domain. Besides the extraction of dominant frequencies containing rich feature information is successfully accomplished after numerous experiments. The other is novel footstep features. Seven simple but effective features are developed and assessed based on dominant frequencies. A SVM is utilized as classifier, exactly identifying which person a FE belongs to. 92.41% precision, 91.3% recall and 91.85% ${F}_{1}$ on average are obtained in personal recognition experiment. Moreover, our seven features show a best performance in the comparison about our features and some features studied previously, even under various SNR. It is worth mentioning that our original signals are collected in noisy environment to approach real application scenarios, without any polishing such as filtering, amplifying. Good classification results could be acquired with poor signal, which is enough to demonstrate that our features are robust and preferable. Our human recognition scheme only involves a microphone to collect footstep sounds and easy classification method, with the characteristics of small calculation and low experimental cost. [ABSTRACT FROM AUTHOR]

At present, the one-stage detector based on the lightweight model can achieve real-time speed, but the detection performance is challenging. To enhance the discriminability and robustness of the model extraction features and improve the detector's detection performance for small objects, we propose two modules in this work. First, we propose a receptive field enhancement method, referred to as adaptive receptive field fusion (ARFF). It enhances the model's feature representation ability by adaptively learning the fusion weights of different receptive field branches in the receptive field module. Then, we propose an enhanced up-sampling (EU) module to reduce the information loss caused by up-sampling on the feature map. Finally, we assemble ARFF and EU modules on top of YOLO v3 to build a real-time, high-precision and lightweight object detection system referred to as the ARFF-EU network. We achieve a state-of-the-art speed and accuracy trade-off on both the Pascal VOC and MS COCO data sets, reporting 83.6% AP at 37.5 FPS and 42.5% AP at 33.7 FPS, respectively. The experimental results show that our proposed ARFF and EU modules improve the detection performance of the ARFF-EU network and achieve the development of advanced, very deep detectors while maintaining real-time speed. [ABSTRACT FROM AUTHOR]

To achieve higher detection rate and lower false alarm rate in dim and small target detection, this paper proposes a novel approach based on the Local Region Similarity Difference ( LRSD ). Then the regions where the true targets and suspected target areas exist in are extracted. To further eliminate the false target regions, we process the LRSD image with pipeline filter and then the true target is finally obtained. Experimental results on some real-life infrared ( IR ) images with different types of targets verify that the proposed method could effectively prove that the proposed method has satisfying detection effectiveness and robustness. Meanwhile, experimental results demonstrate its superior and reliable detection performance by high detection rate and low false alarm rate. [ABSTRACT FROM AUTHOR]