在复杂道路场景下，车辆目标之间频繁遮挡、车辆目标之间相似的外观、目标整个运动过程中采用静态预设参数都会引起跟踪准确率下降等问题。该文提出了一种基于车辆外观特征和帧间光流的目标跟踪算法。首先，通过YOLOv5算法中的YOLOv5x网络模型获得车辆目标框的位置信息；其次，利用RAFT(recurrent all-pairs field transforms for opticalflow)算法计算当前帧和前一帧之间的光流，并根据得到的位置信息对光流图进行裁剪；最后，在卡尔曼滤波过程中利用帧间光流进行补偿得到更精确的运动状态信息，并利用车辆外观特征和交并比特征完成轨迹匹配。实验结果表明，基于车辆外观特征和帧间光流的目标跟踪算法在MOT16数据集上表现良好，相较于跟踪算法DeepSORT，成功跟踪帧数占比提高了1.6%，跟踪准确度提升了1.3%，跟踪精度提升了0.6%，改进的车辆外观特征提取模型准确率在训练集和验证集上分别提高了1.7%、6.3%。因此，基于高精度的车辆外观特征模型结合关联帧间光流的运动状态信息能够有效实现交通场景下的车辆目标跟踪。
In complex road scenes, frequent occlusions and similar appearances between vehicle targets, coupled with the use of static preset parameters used throughout the entire movement of the targets collectively contribute to a decline in tracking accuracy. This paper proposes an object tracking algorithm based on vehicle appearance features and inter-frame optical flow. Firstly, the position information of the vehicle target frame is obtained through the YOLOv5x network model. Secondly, the optical flow between the current frame and the previous frame is calculated using the RAFT (recurrent all-pairs field transforms for optical flow) algorithm, and the optical flow map is clipped according to the obtained position information. Finally, in the process of Kalman filtering, inter-frame optical flow is used to compensate for more accurate motion state information, while vehicle appearance features and intersection over union (IOU) features are used to complete trajectory matching. Experimental results show that the tracking algorithm correlating inter-frame optical flow performs well on the MOT16 data set. Compared with simple online and realtime tracking with a deep association metric (DeepSORT), mostly tracked trajectories (MT) has increased by 1.6%, multiple object tracking accuracy (MOTA) has increased by 1.3%, and multiple object tracking precision (MOTP) has increased by 0.6%. The accuracy of the improved vehicle appearance feature extraction model has been improved by 1.7% and 6.3% on the training and verification sets, respectively. Consequently, leveraging the high-precision vehicle appearance feature model and motion state information from the associated inter-frame optical flow enables effective vehicle target tracking in traffic scenes.
[1] Chen X, Yan B, Zhu J W, et al. Transformer tracking [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2021: 8122-8131.
[2] Zhang Y F, Wang C Y, Wang X G, et al. FairMOT: on the fairness of detection and reidentification in multiple object tracking [J]. International Journal of Computer Vision, 2021, 129(11): 3069-3087.
[3] Meinhardt T, Kirillov A, Leal-Taixe L, et al. Trackformer: multi-object tracking with transformers [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2022: 8834-8844.
[4] Zeng F G, Dong B, Zhang Y, et al. MOTR: end-to-end multiple-object tracking with transformer [C]//European Conference on Computer Vision, 2022: 659-675.
[5] Cai J, Xu M, Li W, et al. MeMOT: multi-object tracking with memory [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2022: 8080-8090.
[6] Jin J T, Li X W, Li X L, et al. Online multi-object tracking with Siamese network and optical flow [C]//IEEE 5th International Conference on Image, Vision and Computing (ICIVC), 2020: 193-198.
[7] Zhang Y F, Sun P Z, Jiang Y, et al. ByteTrack: multi-object tracking by associating every detection box [C]//European Conference on Computer Vision, 2022: 1-21.
[8] Liang C, Zhang Z, Lu Y, et al. Rethinking the competition between detection and ReID in multiobject tracking [J].IEEE Transactions on Image Processing: a Publication of the IEEE Signal Processing Society, 2022, 31: 3182-3196.
[9] Pang J M, Qiu L L, Li X, et al. Quasi-dense similarity learning for multiple object tracking [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2021: 164-173.
[10] Wang Z D, Zheng L, Liu Y X, et al. Towards real-time multi-object tracking [C]//European Conference on Computer Vision, 2020: 107-122.
[11] Paredes-Vallés F, Scheper K Y W, de Roon G C H E. Unsupervised learning of a hierarchical spiking neural network for optical flow estimation: from events to global motion perception [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2051-2064.
[12] Chaney K, Panagopoulou A, Lee C, et al. Self-supervised optical flow with spiking neural networks and event based sensors [C]//IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021: 5892-5899.
[13] Sun D Q, Yang X D, Liu M Y, et al. PWC-net: CNNs for optical flow using pyramid, warping, and cost volume [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8934-8943
[14] Teed Z, Deng J. RAFT: recurrent all-pairs field transforms for optical flow [C]//European Conference on Computer Vision (ECCV 2020), 2020: 402-419.
[15] Zhou X Y, Yin T W, Koltun V, et al. Global tracking transformers [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2022: 8761-8770.
[16] Bewley A, Ge Z Y, Ott L, et al. Simple online and realtime tracking [C]//2016 IEEE International Conference on Image Processing (ICIP), 2016: 3464-3468.
[17] Wojke N, Bewley A, Paulus D. Simple online and realtime tracking with a deep association metric [C]//IEEE International Conference on Image Processing (ICIP), 2018: 3645-3649.
[18] Milan A, Leal-Taixe L, Reid I, et al. MOT16: a benchmark for multi-object tracking [DB/OL]. 2016[2023-0629]. https://arxiv.org/abs/1603.00831.pdf
