Real-Time Detection for Camera Sensing System : Adaptive Cascade Single-Shot Detector

The real-time object detection has recently achieved a series of exciting effect on several dataset like Pascal VOC and MS COCO. These highly-developed algorithms enable real-time image processing from a camera sensing system to be implemented. This paper proposes an object detection model, which is named Adaptive Cascade Single-Shot detector (ACSD) based on earlier Single-Shot Multi box Detector (SSD). The model makes use of adaptive anchor boxes via training dataset statistic experience and add multi-scale cascade layer to reduce over-fitting due to exponentially vanishing positive samples, besides, we proposed a way using spectral clustering to nuke detection dataset. This model aims to improve the detection accuracy and the processing speed in real-time detection tusks. Experiments include several detection tusks on different datasets while the results demonstrate the improvement of ACSD in terms of effectiveness and efficiency. Working performance of the ACSD is verified by comparing it to several state-of-the-art approaches to guarantee it a promising method for the development of the camera sensing system.