基于Yolo与残差网络算法改进的运动目标检测  

Mobile terminal dynamic background motion target detection based on Yolo and residual network

作　　者：郝骏 郑紫微[1] 孙滋昂 杨洋 HAO-jun;ZHENG Zi-wei;SUN zi-ang;YANG-yang(Institute of Communication Technology,Ningbo University,Ningbo 315211,China)

机构地区：[1]宁波大学通信技术研究所,浙江宁波315211

出　　处：《光电子．激光》2020年第1期81-88,共8页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(60972063);浙江省信息与通信工程重中之重学科开放基金(xkxl1524)资助项目。

摘　　要：针对于夜间低光照情况,下雨,下雾等恶劣情况下运动目标检测效果差的问题,以及室内摄像头在夜晚等情况下目标识别率低的问题,提出一种基于Retinex图像增强算法,Retinex算法是根据物体的颜色对长波红色,中波绿色,短波蓝色的光线的反射能力决定的,并不是反射光的强度绝对值决定原理,来消除入射光分量,从而得到反射图像。同时利用残差网络来改进Yolo网络的运动目标算法的改进,并将改进的算法与图像没有增强的Yolo网络以及图像增强后改进后的Yolo网络,实验效果形成对比。根据实验效果对比,在低照度等恶劣情况下,本文目标检测速度提高了0.2 s以上,目标检测的置信度提高20%以上,目标漏检率降低了50%左右,发现本文提出的算法检测精度更高,检测速度较快。Aiming at the problem of low-light conditions at night,poor detection of moving targets under severe conditions such as rain and fog,and the problem of low target recognition rate of indoor cameras at night,a Retinex image enhancement algorithm based on Retinex algorithm is proposed.It is determined by the reflection ability of the long-wave red,medium-wave green,and short-wave blue light according to the color of the object.It is not the principle of determining the absolute value of the intensity of the reflected light to eliminate the incident light component,thereby obtaining a reflected image.At the same time,the residual network is used to improve the improvement of the moving target algorithm of the Yolo network,and the improved algorithm is compared with the Yolo network without image enhancement and the improved Yolo network after image enhancement.According to the comparison of experimental results,in the case of low illumination and other harsh conditions,the target detection speed is increased by more than 0.2 s,the confidence of target detection is increased by more than 20%,and the target missed detection rate is reduced by about 50%.The detection accuracy of the proposed algorithm is found.Higher,faster detection.

关 键 词：低照度 图像增强 Yolo网络 残差网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]