基于机器视觉的动态环境运动目标智能识别研究

Automatic recognition of moving objects in dynamic environment based on machine vision

运动目标自动识别是当前的一个重要研究课题,当前运动目标识别方法存在耗时长、效率低、准确性差等缺陷,为了解决当前动态环境运动目标识别过程存在的缺陷,提高运动目标识别正确率,提出了基于机器视觉的动态环境运动目标自动识别方法。首先研究了运动目标识别进展,分析运动目标识别效果不理想的原因,然后引入机器视觉技术提取运动目标识别的特征向量,并对特征向量进行归一化处理,最后采用机器学习算法根据特征向量进行运动目标识别的分类器,并在Matlab 2019平台上实现运动目标自动识别实验。静态环境中基于机器视觉的运动目标识别正确率和拒识率分别为96.73%和3.27%,动态环境中识别正确率和拒识率分别为93.8%和6.58%,静态环境中识别耗时最高为3.58 s,动态环境中为5.79 s,相对同类运动目标自动识别方法,本方法的运动目标自动识别效果更优,验证运动目标自动识别方法的优越性。

Automatic recognition of moving objects is an important research topic at present. The current methods of moving target recognition have some defects, such as long time-consuming, low efficiency and poor accuracy. In order to solve the defects in the process of moving target recognition in dynamic environment and improve the accuracy of moving object recognition, the automatic recognition method of moving object in dynamic environment based on machine vision is proposed. Firstly, the progress of moving target recognition is studied, and the reason why the recognition effect is not ideal is analyzed. Then the machine vision technology is introduced to extract the feature vector of the moving target recognition, and the feature vector is normalized. Finally, the machine learning algorithm is used to classify the moving target recognition according to the feature vector, and the classifier is implemented in MATLAB. The experiment of automatic recognition of moving target is realized on the platform of 2019. In the static environment, the recognition accuracy and rejection rate based on machine vision are 96.73% and 3.27% respectively, while in the dynamic environment, the recognition accuracy and rejection rate are 93.8% and 6.58% respectively. The recognition time in the static environment is the highest, which is 3.58 s, and in the dynamic environment, which is 5.79 s. Compared with the automatic recognition methods of similar moving objects, the effect of automatic recognition of moving objects in this paper is better, which verifies the superiority of the automatic recognition method of moving objects in this paper.