›› 2017, Vol. 23 ›› Issue (第11): 2382-2391.DOI: 10.13196/j.cims.2017.11.006

Previous Articles     Next Articles

Intelligent robot object detection algorithm based on spatial pyramid and integrated features

  

  • Online:2017-11-30 Published:2017-11-30
  • Supported by:
    Project supported by the National Natural Secience Foundation,China(No.61572438),the Open Fund of the Key Laboratory for Metallurgical Equipment and Control of Ministry of Education in Wuhan  University of Science and Technology,China(No.2017B04).

基于空间金字塔和特征集成的智能机器人目标检测算法

王万良,朱炎亮,王铮,屠海龙   

  1. 浙江工业大学计算机科学与技术学院
  • 基金资助:
    国家自然科学基金资助项目(61572438);武汉科技大学冶金装备及其控制教育部重点实验室开放基金资助项目(2017B04)。

Abstract: With the accelerated study of artificial intelligence in recent years,robots had become research hotspot due to complex demand on recognition,detection and control.By taking NAO soccer robots as the example,aiming at the difficulty of object recognition in competition with complex background,variable illumination and viewpoints,a real-time object detection method based on spatial pyramid and integrated features was proposed.To address the problem caused by different sights,the multi-scale detection was introduced by constructing Gaussian spatial pyramid space through the original image.Furthermore,a dual-channel serial detection framework was proposed,and HOG-PCA basic detector with linear support vector machine which had fast speed and low undetected rate were used to make preliminary detection;an advanced RGB-SIFT-PCA/BOVW detector and random forests with high precision and low error rate were used for secondary screening.Non-maximum suppression algorithm was adopted to remove redundant bounding boxes.Experimental results demonstrated that the proposed method achieved high robustness and real-time performance in intelligent robot object recognition task.

Key words: intelligent robot, object detection, spatial pyramid, integrated features, dual-channel serial detection

摘要: 随着人工智能研究的不断升温,机器人以其对识别、检测、控制的复杂需求逐渐成为研究热点。以NAO足球机器人为例,针对比赛球场背景复杂、光照视角多变、造成目标识别困难的问题,提出一种基于空间金字塔和特征集成的目标实时检测算法。算法引入多尺度检测,通过对原始图像构建高斯金字塔空间,解决了不同视距下目标检测的难点。提出双通道串行特征集成框架,利用计算速度快、漏检率低的梯度方向直方图特征基础检测器和线性支持向量机做初步检测,采用识别精度高、错检率低的三通道尺度不变特征转换描述子改进检测器和随机森林做二次筛选,然后使用非极大值抑制算法去除冗余标定框。实验结果表明,该方法在智能机器人目标识别任务上具有很高的鲁棒性和实时性。

关键词: 智能机器人, 目标检测, 空间金字塔, 特征集成, 双通道串行检测

CLC Number: