Small target pedestrian detection system based on improved YOLO algorithm

Abstract

The invention discloses a small target pedestrian detection system based on an improved YOLO algorithm, and the system comprises a Jetson Nano embedded development board which is used for building a pedestrian detection data set, carrying out the k-means + + clustering analysis of a real target frame in a data set label, and obtaining K types of different prior frame sizes; constructing an improved YOLOv4 model, taking a YOLOv4 Neck layer structure as a basis, and replacing the position of a quintic convolution module in an original feature pyramid with a Dense Net module; each Dense Net module comprises two DCBL modules, and an input feature is sequentially connected with the two DCBL modules through nonlinear variation; meanwhile, the input feature is respectively connected with the outputs of the two DCBL modules in a jumping manner, and the output of the first DCBL module is also connected to the output of the second DCBL module in a jumping manner; and the structure of each DCBL module is a BN-Relu-1 * 1 convolution block to a BN-Relu-3 * 3 convolution block. The low-cost deployment requirement of the mobile terminal is met, and the requirements that the accuracy rate is 85% or above and the detection frame rate is 30 fps or above can be met under different illumination conditions.

Description

technical field [0001] The invention relates to the technical field of pedestrian detection or image processing, in particular to a small target pedestrian detection system based on the improved YOLO algorithm. Background technique [0002] As one of the important research fields of computer vision, pedestrian detection is widely used in intelligent video surveillance, intelligent robot, vehicle assisted driving and data transaction. The use of deep convolutions to build object detection networks has had a profound impact on the field of pedestrian detection. RCNN and its improved series of models classify and regress frames on the basis of candidate suggestion frames, which is of great help to improve accuracy, but the detection speed is reduced. One-stage algorithms such as SSD and YOLO series complete the target classification and positioning tasks at the same time, which greatly improves the detection speed, but there are still certain problems in terms of deployment an...

AI Technical Summary

Problems solved by technology

[0003] At present, there are still deficiencies in multi-target pedestrian detection in the blind area of ​​heavy trucks in mining areas and mobile terminals such as cranes in construction environments: 1) Due to the large proportion of small-scale pedestrians in the recognition environment, pedestrians have different postures and complex backgrounds These factors, such as degree, will affect the accuracy of the small target pedestrian detection algorithm; 2) and because the deep learning model generally has a complex structure, at the same time, it brings a large number of parameters, in real-time pedestrian detection, it is impossible to Better meet the required requirements, the model parameters are large, and it is difficult to detect real-time pedestrians with high accuracy

Images