Multi-target detection method and device and mobile terminal

Abstract

The invention provides a multi-target detection method. The multi-target detection method comprises the steps that a preprocessing module carries out convolution and pooling operation on a to-be-detected image; The first dense connection module performs convolution operation on the output of the preprocessing module and then performs series operation with the output of the preprocessing module; The first transition module performs convolution and pooling operation on the output of the first dense connection module; The second dense connection module performs convolution operation on the outputof the first transition module and then performs series operation with the output of the first transition module; The second transition module carries out convolution operation on the output of the second dense connection module; The third transition module performs pooling and convolution operation on the output of the first transition module; The extraction feature layer carries out convolutionoperation and residual module processing on the output of the first transition module and the series connection result of the output of the second transition module and the output of the third transition module; The prediction layer processes the output of the extraction feature layer and decodes the predicted target position; And the non-extreme value suppression module carries out post-processing on the output of the prediction layer.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a multi-target detection method, device and mobile terminal. Background technique [0002] Target detection is the core problem in the field of computer vision. The main purpose is to analyze image or video information to determine whether there are certain objects (such as faces, pedestrians, cars, etc.), and if so, to give the specific location of these objects . Object detection technology can be widely used in security monitoring, automatic driving, human-computer interaction and other fields, and is the prerequisite for subsequent high-level tasks such as behavior analysis and semantic analysis. [0003] There are many methods of target detection, the most influential traditional methods are based on the deformation model (Deformable Part-based Model, DPM) and self-boosting cascade model (AdaBoost Cascaded Model). The former is successfully used in pedestri...

AI Technical Summary

Problems solved by technology

The shortcomings of these methods can be classified into the following three points: 1. Training the classification model on the huge ImageNet dataset is time-consuming and labor-intensive; 2. The transfer learning (Transfer Learning) model that relies on the pre-trained model for fine-tuning has certain limitations. 3. The use of pre-trained models limits the design space of the model architecture, which is not conducive to the design of flexible and efficient detection models

Images