Improved Centernet complex environment target detection method

Abstract

The invention discloses an improved complex environment target detection method for the Centernet, which is applied to the field of computer vision and aims at solving the problem of low detection performance in special scenes such as crowded, fuzzy and insufficient light in the prior art. According to the method, on the basis of the CenterNet model, the generation rule of the heat map is redefined to adapt to the change of the aspect ratio of the target; meanwhile, an original branch network for predicting the width and the height is adjusted to use multi-output prediction based on a distribution probability model, width and height information is obtained through integration, and the problem of fuzzy boundary of a prediction frame in a crowded scene is relieved; and the heat map is processed based on a maximum suppression substitution scheme of multi-core maximum filtering, so that the confidence coefficient of a false detection frame is reduced to a certain extent while the optimal detection frame is reserved, and the overall precision is improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a complex scene target detection technology. Background technique [0002] Object detection is an important research branch in the field of computer vision. The problem to be solved by target detection can be simply defined as "what target is where". At present, the algorithm structure of the network model can be divided into "two-stage" and "one-stage" target detection algorithms. The difference is that the former is detected in two steps. Target location: First, generate a possible target frame based on the Region Proposal Network (RPN, Region Proposal Network), and then classify the target frame, that is, transform the detection problem into a classification problem; the latter is directly regressed from the convolutional layer features of the image. The target box is to transform the detection problem into a regression problem. The previous two-stage network can o...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above-mentioned technical problems, the present invention proposes an improved Centernet object detection method in a complex environment, which uses a probability-based form to predict the probability distribution of the width and height of the target frame, and determines the width and height by integral instead of directly predicting the width and height information. It can give a more stable prediction of width and height. In addition, the newly defined heat map rule reflects the shape of the target, and the proposed post-processing method of multi-core maximum filtering alleviates the problem of repeated detection of large targets in the original model to a certain extent.

Images