Scene-Independent Motion Pattern Segmentation (MPS) in crowded video scenes using Spatio-Angular Density-based Clustering (SADC)

About the Dataset

• The dataset used for this work (CUHK Crowd Dataset) is obtained from: https://amandajshao.github.io/projects/CUHKcrowd_files/cuhk_crowd_dataset.htm

• The CUHK dataset contains a total of 474 video clips. Among these, only 300 video clips have ground truth for MPS.

• The file CUHK_info.xlsx (obtained from the above mentioned dataset link) contains the details of each video clip, where we have highlighted (in yellow-color) the 300 video clips used for MPS.

• For this work, we have categorized the 300 video clips into 8 different categories:

  cross walk, escalator, market, mass movement, public walkway, shopping mall, station, and street.

• We also categorize the scenes into:

  structured-unstructured and indoor-outdoor.

• The file CUHK_300_category_info.xlsx contains all the category-wise details. The last column in this file can be used as a mapping to scenes in the original file CUHK_info.xlsx

• For evaluation and comparison of our work with frame-by-frame clustering techniques (CM & CT -refer our paper), we use the frame number mentioned in the column "GT end frame no." of the CUHK_300_category_info.xlsx file.

Instructions to run the code

• Just run the file named "sadc.m". Here's an example of how to run it:

[clust_id,gt_id] = sadc(91,20,15); % syntax : sadc(scene_no,eps_threshold,aplha_threshold)

• Please refer to the code for more details regarding the parameters.

• The folders "ground_truth_grDetect" and "CUHKcrowd_dataset_imgTrk" (extract the CUHKcrowd_dataset_imgTrk.zip file to get this folder) are created only for testing purposes. It contains info. of only 1 scene (scene no. 91 of the CUHK dataset).

• Please download the original CUHK dataset to get all the data within these folders.

Cite

If you use our code or the scene-wise data we have created, please cite our paper:

A. K. Pai, A. K. Karunakar and U. Raghavendra, "Scene-Independent Motion Pattern Segmentation in Crowded Video Scenes Using Spatio-Angular Density-Based Clustering" in IEEE Access, vol. 8, pp. 145984-145994, 2020.