Abstract: As we evolve in the field of technology, the applications of the same in our day-to-day life have improved our lives. But this ease comes at a cost since, as the number of appliances used exponentially increases in our society, so does our demand for energy. This energy is usually obtained after destructive emissions are released into the environment, and the wastage of said energy can also be seen in several areas. A survey shows that a plugged-in mobile charger, when not used, consumes 0.1 to 0.5 watts per hour, costing ~15 rupees per day. Such wastage can be expected from other devices as well. Therefore, there is a scope for efficient energy management and, in recent times, it can be done more effectively using automation systems that eliminate the need for human interactions with devices or tools as much as possible. These systems may require the additional fitting of hardware such as sensors that will incur supplementary energy consumption and overhead costs for maintenance. They may also offer low-coverage for detection. Collaterally, it is worthwhile to note that the need for closed-circuit television (CCTV) installations is a basic necessity in several areas of the society and cannot be compromised to conserve energy and yet, we can utilize them to make a region of interest (ROI) energy efficient. This can be achieved through rationing the power consumption of other devices by inferring from an intelligible detection of objects and activities observed by a camera. It is made possible through a technique called Computer Vision (CV). CV, though reducing the workload of building setups for recognition, is a computationally exhaustive technique that requires hardware support to function with an accelerated performance for object detection in real-time. Thus, this work details the different methods available to detect objects and the techniques that can be employed to accelerate the performance of a low power consuming detection system. Graphics Processing Unit (GPU) acceleration and Edge computing are also discussed as a way to offer additional support to CV computation. The advantages and specific drawbacks of each method are also elaborated.
Keywords: Closed Circuit Television (CCTV), Region of Interest (ROI), Computer Vision (CV), Region Based Convolutional Neural Network (R-CNN), You Only Look Once (YOLO), Graphics Processing Unit (GPU), Edge Computing.
|
DOI:
10.17148/IJARCCE.2022.11763
