Border Management Using Internet of Things

Abstract

Single shot multi-box detector is the quickest approach that employs a single layer of neural network to recognize items from an image and video. The system employs numerous photos to recognize items and label them with the appropriate class label. The suggested approach is used in conjunction with a multilayer convolutional neural network to improve computational performance by using a greater number of default boxes. This results in more accurate identification. Raspberry Pi, Pi cam and Raspbian operating system have been used in this work. In this paper, an ultrasonic sensor has been used to measure the distance of the object. The camera starts livestream when the object is closer than a threshold distance. A standard dataset named coco.names is used which can detect 80 different types of objects. The livestream video output is downscaled to an extent where the single shot multi-box detector algorithm cannot detect and recognize the objects. Next, a pre-trained faster super resolution convolutional neural network is used for super resolution to upscale the livestream video. As a result, single shot multi-box detector algorithm can detect and recognize the objects present in the video. In this work, the single shot multi-box detector algorithm has also been tested for detection of an object after enhancing the quality of a low-resolution livestream video using super resolution pretrained model. © 2022 IEEE.