基于目标检测结果的轮廓及颜色识别研究

Research on Object Contours Extraction and Color Recognition Based on Object Detection Result

基于深度学习方法给出的目标矩形框检测结果,针对实时目标轮廓提取和颜色识别问题,提出了一种基于边缘提取和形态学操作的方法.首先通过Canny边缘检测算法提取图像大致轮廓,应用多次形态学闭操作将目标主体与背景、噪声等加以区分,找出最大轮廓即目标轮廓,然后利用目标轮廓所包含的区域,在HSI颜色空间中完成目标颜色的统计和识别,并采用真实场景中的无人机、小汽车和人的图像来进行实验验证.实验结果表明,所提出的方法相比纯粹基于深度神经网络的方法在效率上有较大提升,相比纯粹的底层图像处理方法在精度上有较大提高,既保证了实时性,又确保了较高的精度.

Based on the detection results of the object rectangular frame given by the deep learning method,this paper proposes a method based on edge extraction and morphological operation for the real-time object contour extraction and color recognition.This paper makes use of the Canny edge detection algorithm to extract the rough contour of the image,and then distinguishes the object from the noise and the background through multiple morphological operation of closing.Then the maximum contour is found to be the object contour.The statistics and recognition of the object color in HSI color space is completed by using the area included in the object contour.The UAV,vehicle and human images in real scenes are used to verify the algorithm in this paper.The experimental results show that,compared with the direct application of Canny algorithm and the deep neural network-based method,our method not only guarantees the running speed of the algorithm in the real-time scene,but also ensures a higher accuracy.