Method for detecting target in vertical ladder based on cloud server

Abstract

The invention relates to a method for detecting a target in a vertical elevator based on a cloud server. The method comprises the following steps: a, acquiring image data in a lift car through a liftcar camera of the vertical elevator according to a picture capturing condition; b, writing the image data into a message management module according to node information and a screenshot condition, andstoring the image data into a cloud memory; c, reading the image data from the cloud memory through the message of the consumption message management module and providing the image data to the cloudprocessor; and d, after the cloud processor carries out batch image preprocessing on the image data, using a deep learning detection model deployed on the cloud processor to carry out accelerated reasoning operation, and finally, feeding a detection result back to the data platform. According to the method, the calculation amount of a detection model is reduced by accelerating preprocessing such as image encoding and decoding and scale transformation and replacing and cutting a backbone network, the reasoning time is reduced by using TensorRT acceleration and the like, the performance of a target detection algorithm is optimized to the maximum extent, and the processing capacity of an image in unit time is improved.

Description

Technical field [0001] The invention relates to the field of computer vision, in particular to a method for detecting a target in a straight ladder based on a cloud server. Background technique [0002] As a new research direction in the field of machine learning, deep learning has achieved many results in image recognition, speech recognition, natural language processing and other related fields. By learning the rules and expression levels of sample data, the machine can imitate human audiovisual and thinking, has the ability to analyze and learn, and can accurately recognize images, sounds, texts and other data. For example, for target detection tasks, firstly, sample data needs to be collected, sample information is labeled, and a model is obtained after training and testing. In the final application stage, the trained model needs to be deployed to client machines, servers or embedded devices as a production environment. The general production environment has clear performanc...

AI Technical Summary

Problems solved by technology

If the image is large and the detection model network is deep, the image preprocessing time and algorithm reasoning time will be time-consuming, and the data processing will be slow.

Images