Multi-class target detection method and system

Abstract

The invention discloses a multi-class target detection method, and the method comprises the steps: searching a first class target and a second class target in a to-be-detected image, and obtaining a positioning frame; classifying the images in the positioning frame, and obtaining a background probability and a target probability; and filtering and classifying the uncertain images in the positioning frame, and obtaining a detection result. Compared with the prior art, the invention has the beneficial effects that due to the fact that the lightweight convolutional neural network is adopted, theimage detection function does not need to be deployed to the cloud and directly runs on embedded hardware in the camera, network transmission delay is avoided, and meanwhile, the purchase cost of a user is reduced. And besides, human-shaped pet positioning and classification are separately processed in time sequence, and a comparison filtering mechanism is adopted, so that false detection is finally reduced to a greater extent.

Description

technical field [0001] The present application relates to the field of image detection, in particular, to a method for multi-category object detection. Background technique [0002] Indoor human and pet detection refers to the collection of indoor pictures of residents through cameras, and the positioning and classification of people and pets appearing in the pictures through image detection technology. [0003] With the advancement of technology and the improvement of people's living standards, home monitoring cameras have gradually entered thousands of households. Through mobile apps, people can actively check the situation at home at any time, and have a certain role in caring for the elderly, infants or pets. Furthermore, there are currently merchants that provide image detection functions to analyze the images collected by surveillance cameras, and push the detected pictures of people or pets to users through the app, so that users can grasp information in real time. ...

AI Technical Summary

Problems solved by technology

[0004] However, there are the following disadvantages in the prior art: 1. This image detection function is generally implemented based on a convolutional neural network (CNN), which requires a large amount of calculation, and is deployed on a dedicated graphics processing unit (GPU) in the cloud as a value-added The service needs to be purchased by the user for an additional fee

2. From camera collection, transmission, cloud detection and processing, to the final app message push, there is a large delay, so strictly speaking, this method is not real-time, and users cannot obtain the maximum utility

3. Due to the complex indoor environment and poor lighting conditions, the current image detection algorithm is not robust enough, and a certain degree of false detection will occur

Images