摘要
传统方法中的IDT,对于行为识别效果最好,但其中的光流计算太缓慢。将FlowNet2.0引入光流估计替代Farneback方法,在保持性能的同时,计算速度提高近7倍。LIBLINEAR的使用,提升了SVM的速度,在UCF-101、KTH、Weizmann数据集上均取得较好效果,IDT方法得到进一步优化。
IDT in the traditional method has the best effect for behavior recognition,but the optical flow calculation is too slow.In this study,FlowNet2.0 was introduced into the estimation instead of the Farneback method,which improved the calculation speed by nearly 7 times while maintaining the performance.The use of LIBLINEAR greatly improved the speed of SVM and achieved good results on UCF-101,KTH and Weizmann datasets,which further optimized IDT method.
出处
《工业控制计算机》
2020年第4期46-48,共3页
Industrial Control Computer
