Method, system, medium and device for improving pedestrian recognition speed in restricted area scene

Abstract

The invention belongs to the technical field of target recognition, and specifically provides a method, system, medium and device for improving the speed of pedestrian recognition in a restricted area scene. The present invention aims to solve the problem that the computing power of the existing embedded chips cannot meet the real-time requirements of the application when facing a large-scale CNN model. For this purpose, the present invention reduces the recognition class of YOLO model to recognize m class, and the channel number of the last layer is modified to (m+5)×3 layers; Perform the first training in the model; perform sparse training and regular channel pruning on the YOLO model completed for the first training, so that the number of channels in other layers except the last layer is 2 n , to get the revised YOLO model; use the DNNDK reasoning framework to generate the code of the YOLO model; input the code into the recognition device with an FPGA chip to run, so as to improve the pedestrian recognition speed of the recognition device in the restricted area scene.

Description

Technical field [0001] The present invention belongs to the field of target recognition, and provides a method, system, medium, and apparatus for enhancing the velocity of pedestrian recognition in the restricted area. Background technique [0002] It is important to target the target intrusion detection. The target can be pedestrians or animals. The penalty area can be a railway, high pressure zone, an animal protection area, and the like. The restricted area is taken as an example, and the goal is taken as an example of a pedestrian. At present, the pedestrian detection algorithm based on convolutional neural network (CNN) is a more popular research direction. As the number of network layers is constantly deepening, the algorithm accuracy has increased significantly, and huge amounts of calculation are also introduced. Pedestrian detection applications are typically deployed directly to embedded ends, and the chip of embedded ends is difficult to meet the real-time demand for t...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned problems in the prior art, that is, in order to solve the problem that the computing power of the existing embedded-end chip is difficult to meet the real-time requirements of the application when facing a large-scale CNN model, the present invention provides a A method for improving the speed of pedestrian recognition in a restricted area scene, including:

Images