PLCFishMOT: multiple fish fry tracking utilizing particle filtering and attention mechanism

被引:0
|
作者
Tan, Huachao [1 ]
Cheng, Yuan [2 ]
Liu, Dan [1 ]
Yuan, Guihong [1 ]
Jiang, Yanbo [1 ]
Gao, Hongyong [1 ]
Bi, Hai [3 ]
机构
[1] Dalian Ocean Univ, Coll Informat Engn, Dalian 116000, Peoples R China
[2] Dalian Univ Technol, Ningbo Inst, Ningbo 315000, Peoples R China
[3] Hangzhou Yunxi Smart Vis Technol Co Ltd, Dalian 116000, Peoples R China
关键词
Fish tracking; Multiple object tracking; Particle filtering; Attention mechanism; Minimum cosine distance; Intersection over Union; MODEL;
D O I
10.1007/s10499-024-01713-y
中图分类号
S9 [水产、渔业];
学科分类号
0908 ;
摘要
The task of multi-object tracking of fish fry poses significant challenges, as the majority of the fish fry individuals exhibit highly similar appearances, and the feature distinctions between individual targets are not readily apparent. Consequently, fish tracking algorithms relying primarily on appearance-based features for data association often suffer from low accuracy and poor robustness. To address the challenges inherent in multi-object tracking of fish fry, this study presents an improved DeepSort-based algorithm, dubbed PLCFishMOT, designed specifically for enhanced performance in this domain. Furthermore, the fish fry trajectories may exhibit nonlinear characteristics due to external perturbations. To address this, the original Kalman filtering method has been replaced with a particle filtering approach, which is more suitable for handling nonlinear and non-Gaussian problems. This modification serves to enhance the accuracy of the trajectory prediction process. To further bolster the accuracy of the data association process, the proposed framework incorporates a large separable kernel attention mechanism into the original feature extraction network. This mechanism leverages convolutional kernels of varying sizes to extract target features with differing receptive field dimensions, thereby enhancing the overall effectiveness of the feature representation. The proposed approach effectively addresses the challenge of incorrect ID assignment, which can arise due to the close parallel swimming patterns exhibited by the fish fry. This is achieved by leveraging the cosine angle value between the fry detection frame and the trajectory frame as a discriminating factor. The experimental evaluation of the proposed algorithm on an open-source video dataset demonstrates its strong performance, with the algorithm achieving an IDF1 score of 75.8%, a MOTA score of 98.1%, and IDs is 10, respectively. Furthermore, to assess the generalization capabilities of the proposed approach, validation experiments were conducted using a fish fry video dataset captured in real-world aquaculture scenarios. The experimental results demonstrate that the PLCFishMOT algorithm achieves the best tracking performance compared to other advanced multi-object tracking algorithms.
引用
收藏
页数:25
相关论文
共 50 条
  • [1] Tracking multiple objects with particle filtering
    Hue, C
    Le Cadre, JP
    Pérez, P
    IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS, 2002, 38 (03) : 791 - 812
  • [2] Target tracking by multiple particle filtering
    Bugallo, Monica F.
    Lu, Ting
    Djuric, Petar M.
    2007 IEEE AEROSPACE CONFERENCE, VOLS 1-9, 2007, : 1882 - 1888
  • [3] Optimal particle allocation in particle filtering for multiple object tracking
    Pan, Pan
    Schonfeld, Dan
    2007 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING, VOLS 1-7, 2007, : 61 - 64
  • [4] Fish tracking by combining motion based segmentation and particle filtering
    Bichot, E
    Mascarilla, L
    Courtellemont, P
    VISUAL COMMUNICATIONS AND IMAGE PROCESSING 2006, PTS 1 AND 2, 2006, 6077
  • [5] Resource management in particle filtering for multiple object tracking
    Pan, Pan
    Schonfeld, Dan
    VISUAL COMMUNICATIONS AND IMAGE PROCESSING 2008, PTS 1 AND 2, 2008, 6822
  • [6] Target tracking using multiple auxiliary particle filtering
    Ubeda-Medina, Luis
    Garcia-Fernandez, Angel F.
    Grajal, Jesus
    2017 20TH INTERNATIONAL CONFERENCE ON INFORMATION FUSION (FUSION), 2017, : 9 - 16
  • [7] Efficient particle filtering for multiple target tracking with application to tracking in structured images
    Maskell, S
    Rollason, M
    Gordon, N
    Salmond, D
    IMAGE AND VISION COMPUTING, 2003, 21 (10) : 931 - 939
  • [8] Efficient particle filtering for multiple target tracking with application to tracking in structured images
    Maskell, S
    Rollason, M
    Gordon, N
    Salmond, D
    SIGNAL AND DATA PROCESSING OF SMALL TARGETS 2002, 2002, 4728 : 251 - 262
  • [9] OBJECT TRACKING BASED ON PARTICLE FILTERING WITH MULTIPLE APPEARANCE MODELS
    Widynski, Nicolas
    Aldea, Emanuel
    Dubuisson, Severine
    Bloch, Isabelle
    VISAPP 2011: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON COMPUTER VISION THEORY AND APPLICATIONS, 2011, : 604 - 609
  • [10] Particle filtering with multiple cues for object tracking in video sequences
    Brasnett, P
    Mihaylova, L
    Canagarajah, N
    Bull, D
    IMAGE AND VIDEO COMMUNICATIONS AND PROCESSING 2005, PTS 1 AND 2, 2005, 5685 : 430 - 441