1.深度學習的目標跟蹤算法綜述_盧湖川
2.Large Margin Object Tracking with Circulant Feature Maps(APCE)
1.高峯檢測
2.多置信度更新
(LMCF置信度指標是平均峯值相關能量(average peak-to correlation energy, APCE))
(置信度指標還有MOSSE中的峯值旁瓣比(Peak to Sidelobe Ratio, PSR))
(CSR-DCF的空域可靠性)
3.fully-convolutional siamese networks for object tracking (siamesefc)
5個尺度(58 FPS) 3個尺度(86FPS)
1.數據及大(4500個視頻)
2.卷積特徵
4.搜索區域大(4倍)
//srdcf(速度慢)
LMCF
BACF
staple
eco
siamesefc
yolo
fastercnn
mdnet
goturn
maskrcnn
總結
//srdcf(速度慢)
ICCV,ECCV,CVPR
HuChuan Lu
特徵選擇(HOG,CNN,CN)
分類器(SVM,嶺迴歸,Deep Netural NetWorks)
模型更新(Fixed interval,High-condifience strategy)
入門
1.Uderstading and diagnoising visual tracking systems
2.KCF
3.Learning multi-domain convotional netural networks for visual tracking(Mdnet)
網址: https://github.com/foolwood/benchmark_results