An Optimization Method for Face Detection Based on Deep Convolutional Cascaded Networks

Abstract

The present invention provides a face detection optimization method based on a deep convolutional cascaded network, which specifically includes the following steps: using the deep cascaded network to detect areas where human faces may appear, that is, hot spots; updating the hot spots, Set all the areas not in the hot zone to zero; perform data sparse compression on the obtained whole image. The invention greatly reduces the amount of redundant computation and improves the operating efficiency of the algorithm. The efficiency of the CNN network face detection algorithm on the front-end embedded platform can be increased by 20-30%, and the method of the present invention will not lose the detection accuracy, providing a better solution for the faster operation of the deep convolutional network on the front-end The scheme laid the foundation for the large-scale application of the network in the later stage.

Description

technical field [0001] The invention belongs to the technical field of automatic detection, and in particular relates to a face detection optimization method based on a deep convolution cascade network. Background technique [0002] With the rapid development of deep learning technology, great breakthroughs have been made in the fields of target detection and recognition. Wide and deep networks have brought better results, but the amount of calculation is also astonishing. How to make the CNN network Being able to run on an embedded system is a problem we are currently facing. The general approach to solve this problem is: 1. Design a lighter CNN network to reduce the amount of calculation, but small networks generally sacrifice accuracy. 2. According to different embedded platforms, special instructions are used to accelerate, and it takes time and effort to write and debug. 3. The hard core with CNN network is used to realize it. Although the efficiency is not a problem,...

AI Technical Summary

Problems solved by technology

The general approach to solve this problem is: 1. Design a lighter CNN network to reduce the amount of calculation, but small networks generally sacrifice accuracy.

2. According to different embedded platforms, special instructions are used to accelerate, and it takes time and effort to write and debug

3. The hard core with CNN network is used to realize it. Although the efficiency is not a problem, the cost is relatively high

Images