End-to-end object detection method based on convolutional neural network

Abstract

The invention discloses an end-to-end object detection method based on a convolutional neural network. The method includes the following steps: (1) on the basis of a classic basic network structure, removing the last fully-connected layer of the classic basic network, and adding an extra layer so as to establish a convolutional neural network model; (2) randomly selecting, from an original training data set, an original image, obtaining an amplified image through data amplification, and obtaining the position and frame of an object image block, randomly selected from the original image, in the amplified image; (3) through the position and frame of the object image block in the amplified image obtained in the step (2), conducting regression on the convolutional neural network model in the step (2) to obtain model parameters, and thus obtaining a trained convolutional neural network model; and (4) utilizing the trained convolutional neural network model to detect a bounding box and a classification of an object in an image to be detected. Direct regress of a central point coordinate, the width, the height and the classification of the object is adopted by the method. Compared with similar methods, the method has a significant advantage in terms of speed.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically, relates to an end-to-end object detection method based on a convolutional neural network. Background technique [0002] Target detection is a basic task in computer vision. It can be used in many common projects in reality, such as pedestrian detection, vehicle detection, target tracking and preprocessing in image retrieval. Doing a good job in target detection is very helpful for some higher-level tasks. Most of the existing target detection algorithms are algorithms with high complexity, and few can achieve real-time. Therefore, developing a set of high-precision and fast detection algorithms has always been a difficult problem in computer vision. Contents of the invention [0003] In view of the above defects or improvement needs of the prior art, the present invention provides an end-to-end object detection method based on a convolutional neural network, which has high...

AI Technical Summary

Problems solved by technology

Most of the existing target detection algorithms are algorithms with high complexity, which rarely achieve real-time performance. Therefore, developing a set of high-precision and fast detection algorithms has always been a difficult problem in computer vision.

Images