An Object Detection Method Based on Semantic Segmentation Enhancement

Abstract

The invention relates to an object detection method based on semantic segmentation enhancement, including: preparing marked images; image set division; designing a deep convolutional neural network structure based on semantic segmentation enhancement to be suitable for object detection; The deep convolutional neural network consists of three main parts: the backbone subnetwork, the segmentation subnetwork and the detection subnetwork. The backbone subnetwork is used to extract general features of the image, which are category-independent features; the segmentation subnetwork is based on the backbone subnetwork extraction Based on the features, further extract the features of semantic segmentation and predict the heat map of each type of object segmentation; use the heat map of each type of object as the prior knowledge of this class, and fuse it with the features extracted by the detection sub-network, Then generate category-related features, each type of object has a feature of the corresponding category, which significantly reflects the characteristics of this type of object; model training.

Description

technical field [0001] The invention relates to a high-performance object detection method in the field of computer vision, in particular to a method for image object detection using a deep learning method. Background technique [0002] As a key technology in the development of artificial intelligence, deep learning technology has been widely used in many fields such as intelligent monitoring, human-computer interaction, assisted driving, and automatic driving to realize real-time detection and monitoring of people, cars, and other objects in the scene. identify. As an important implementation method in deep learning technology, deep convolutional neural network has achieved remarkable results in object detection tasks. [0003] Taking the autonomous driving system as an example, such as figure 1 As shown, in the object detection task, the video / image in the real scene is first captured by the vehicle camera; further, the video / image captured by the camera is input into th...

AI Technical Summary

Problems solved by technology

[0012] The purpose of the present invention is to overcome the problem of low detection accuracy of existing object detection algorithms based on deep convolutional neural networks, and propose a deep convolutional neural network object detection method based on semantic segmentation enhancement, which can effectively improve object detection The accuracy further promotes the application of object detection in many fields

Images