With the development of human society, the shortcomings of the existing transportation system become increasingly prominent, so people hope to use advanced technology to achieve intelligent transportation. However, the recognition rate of most methods of detecting video vehicles is too low and the process is complicated. This paper uses machine learning theory to design a variety of pattern classifiers, including Adaboost, SVM, RF, and SVR algorithms, to classify vehicles. Support vector regression (SVR) is a support vector regression algorithm based on the basic principles of support vector machine (SVM) and then generalized to the regression problem. This paper proposes a short-term traffic flow prediction model based on SVR and optimizes SVM parameters to form an improved SVR short-term traffic flow prediction model. It can be obtained from experiments that the classification error rate of support vector regression (SVR) is the lowest (3.22%). According to the prediction of morning and night peak hours, this paper concludes that the MAPE of SVR is reduced by 19.94% and 42.86%, respectively, and the RMSE is reduced by 29.71% and 47.22%, respectively. Experiments show that the improved algorithm can obtain the optimal parameter combination of SVR faster and better and can effectively improve the accuracy of traffic flow prediction. The target tracking pedestrian counting method proposed in this paper has significantly improved the counting accuracy. The calculation of HOG features can be further expanded, such as the selection of neighborhoods when calculating HOG features, and finally a more efficient pedestrian counting framework is implemented.

随着人类社会的发展，现有交通系统的弊端日益突出，因此人们希望利用先进的技术来实现智能交通。但是，大多数检测视频车辆的方法的识别率都太低，并且过程复杂。本文使用机器学习理论来设计各种模式分类器，包括Adaboost，SVM，RF和SVR算法，以对车辆进行分类。支持向量回归（SVR）是一种基于支持向量机（SVM）基本原理的支持向量回归算法，然后将其推广到回归问题。提出了一种基于SVR的短期交通流量预测模型，优化了SVM参数，形成了一种改进的SVR短期交通流量预测模型。从实验中可以得出，支持向量回归（SVR）的分类错误率最低（3.22％）。根据早晚高峰时段的预测，得出结论：SVR的MAPE分别降低了19.94％和42.86％，RMSE分别降低了29.71％和47.22％。实验表明，改进算法可以更快，更好地获得SVR的最优参数组合，可以有效提高交通流量预测的准确性。本文提出的目标跟踪行人计数方法大大提高了计数精度。HOG特征的计算可以进一步扩展，例如在计算HOG特征时选择邻域，最后实现更有效的行人计数框架。