Multi channel video detection technology based on ascend AI processor

DOI:	10.11809/bqzbgcxb2025.01.033
Keywords:	Ascend AI processor; video detection; image segmentation; localization; algorithm deployment
Abstract:	Ascend AI processor is a new localized neural network accelerator specially launched for application scenarios of artificial intelligence. Different from traditional Graphics Processing Units, Ascend AI processor uses DaVinci architecture independently developed by Huawei to design dedicated AI computing units with high computing power and low power consumption, having great advantages in computer vision applications. As one of the key technologies of computer vision field, object detection technology has been widely used in video surveillance, face recognition and so on. Considering that the view range of traditional single channel video detection is limited and can’t cover multiple different scenarios at the same time, this paper proposes a multi channel video detection technology based on Ascend AI processor. The YOLOv5 algorithm was used for object detection. The multi channel video detection application was developed based on Ascend Computer Language. Functional modules such as resource initialization module, data preprocessing module, image segmentation module, model inference module, data post processing module, and video streaming push and pull module were designed. The test and application results show that the proposed technology can simultaneously pull and detect multiple video streams from different video sources, label the target location and category, and save each video detection results or push the stream to the server respectively. It can be deployed in edge devices in practical applications to realize real time detection of multiple videos.
Issue:	Vol. 46 No. 1 (2025)
Published:	2025-01-31
PDF