整合多頂照式魚眼和投射式攝影機之人物追蹤

Combining Projective and Top-View Fisheye Cameras for People Tracking

謝立璿(Li-Hsuan Hsieh)

指導教授 ： 王才沛

國立交通大學/資訊學院/資訊科學與工程研究所/碩士(2019年)

摘要

在影像辨識的相關研究當中，多物件追蹤是常見的一個議題。一般研究都是使用投射式攝影機作為主要影像來源，當場景中人物較多時，容易產生被遮蔽和交錯的情況，導致追蹤效果不佳；少數研究有使用到魚眼攝影機做為影像來源，它的優點在於比較能減少人物遮蔽的情況，但因為魚眼攝影機影像扭曲程度相較更大，若要利用人物的外觀屬性輔以追蹤較有難度。本篇論文整合以上兩種攝影機的優點，以魚眼攝影機做為主要影像來源，輔以投射式攝影機，並提出新的方法，希望能在即時影像上完成人物追蹤。
本篇論文使用和場景模型等比例的3D voxel來儲存前景值資訊，壓縮成2D map後去尋找每一個frame的前景值做為人物的點座標集合，輔以particle filter的方式來更新人物位置，達到追蹤的效果。人物交錯處理方面，使用color histogram搭配人物外觀屬性，提升追蹤系統的穩定性及精確度。

關鍵字

人物追蹤 ； 魚眼追蹤 ； 攝影機追蹤

並列摘要

In the related work of image processing, multiple object tracking is a very common topic. In general research, projective cameras are used as the main source of images. When there are many people in the scene, it is easy to be masked and staggered, resulting in poor tracking performance. A few studies have used fisheye cameras as the source of images. Its advantage is that it can reduce the occlusion. But, due to the image of the fisheye camera is more distorted, it is more difficult to use the appearance attribute of the character to improve the tracking. This paper combines the advantages of the above two cameras, using the fisheye cameras as the main source of images, supplemented by the projective cameras. Also, we refer a new method of tracking, hoping to complete the object tracking in real-time.
This paper uses a 3D voxel, which size is the same as our scene, to store the foreground information, compress it into a 2D map and then find the value of foreground of each frame as the point coordinate set. Supplemented by the particle filter to update the people’s position to improve the tracking. To handle the occlusion problem, our approach is to combine the color histogram and the appearance model. This can improve the stability and accuracy of our tracking system.

並列關鍵字

people tracking ； fisheye tracking ； camera tracking

參考文獻

[1] W. Luo, J. Xing, A. Milan, X. Zhang, W. Liu, X. Zhao, and T.-K. Kim, Multiple Object Tracking: A Literature Review, 2017.

[2] B. Bose, X. Wang and E. Grimson, “Multi-class object tracking algorithm that handles fragmentation and grouping,” Proc. 2007 IEEE Conference on Computer Vision and Pattern Recognition, pp. 1-8, 2007.

[3] W. Hu, X. Li, W. Luo, X. Zhang, S. Maybank and Z. Zhang, “Single and Multiple Object Tracking Using Log-Euclidean Riemannian Subspace and Block-Division Appearance Model,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, pp. 2420-2440, 2012.

[4] B. Cuan, K. Idrissi and C. Garcia, “Deep Siamese Network for Multiple Object Tracking,” Proc. 2018 IEEE 20th International Workshop on Multimedia Signal Processing (MMSP), pp. 1-6, 2018.

[5] Z. Chen et al., “Object Tracking over a Multiple-Camera Network,” Proc. 2015 IEEE International Conference on Multimedia Big Data, pp. 276-279, 2015.