面向复杂背景环境目标的快速识别研究

Research on Fast Recognition of Target for Complex Background

面向复杂背景环境的定位目标的快速识别是野外作业机器人的关键技术,以柑橘为例,研究了自然环境下基于彩色信息的目标定位的识别方法。首先,采用YCbCr颜色模型来分析柑橘彩色图像的颜色和灰度特征,并通过Otsu与FCM分割算法相结合对在不同光照条件下拍摄的彩色目标图像进行分割;然后,利用形态学数学和区域标记消除分割后产生的随机噪声;最后,用凸包算法提取果实形状特征,并通过凸包算法来判定是否为柑橘和是否可采。对500张彩色柑橘图像进行分割,结果表明采用Cr颜色分量和Otsu与FCM算法相结合有效地解决复杂自然光照下的分割问题;对963个柑橘进行了凸包算法识别试验,总体识别率达87.53%。凸包算法对遮挡图像也可进行高效识别,并能快速、准确地提取柑橘目标的质心坐标。

Fast recognition of target for complex background was the key technology of the field robot. Take citrus as ex- ample, the recognition method was researched based on color information in the natural environment. First of all, the YCbCr color model is used to analyze the color and grayscale characteristics of the citrus color image. Color target image in different lighting conditions will be segmented by Otsu and FCM segmentation algorithm;Then, morphological mathe- matics and regional mark eliminate the random noise generated in the segmented; Finally, the convex hull algorithm will extract characteristics of fruit shape, and convex hull algorithm will determine whether citrus can be picked. 500 color cit-rus images are segmented, the results show that the Cr color components and Otsu Combined with FCM algorithm effec-tively solve the segmentation problem under complex natural light; By testing 963 citrus recognition used by the convex hull algorithm, the rate of overall recognition have achieved 86.3 %, and the recognition rate of citrus overlapped exceed over 87.6 percent, this indicated that the shape feature extraction methods based on convex hull algorithm, effectively recognited images of citrus overlapped and blocked and the citrus center of mass coordinates of the target can be quickly and accurately extracted.