Collaborative cameras data

Recorded Sequences and Unreal Engine Project used to evaluate Collaborative Camera Systems

The synthetic sequences recorded to evaluate our work "Distributed multi-target tracking and active perception with mobile camera networks"  (demo video) can be downloaded from here. The dataset includes six sequences captured in two photo-realistic environments built in Unreal Engine 4.27  and designed using our "Framework for Fast Prototyping of Photo-realistic Environments with Multiple Pedestrians" (github):

• Font environment: green and open area where three different sequences have been acquired Font Sparse (5 people), Font Medium (10 people) and Font Busy (15 people).

• Street environment: commercial street where three different sequences have been acquired Street Sparse (5 people), Street Medium (10 people) and Street Busy (15 people).

Each sequence includes records from 3 static cameras and 2 mobile cameras (drones) with a length of 500 frames of 1440x900 pixels.

The structure of each sequence is as follows:

The ground truth of the pedestrians in the image plane is given by the json file "pedetrian_2dGT.json" included in each one of the camera folders. The structure of the ground truth dictionary is:

{'frame_index_0': [{'id': pedestrian_name, 'xmin':xmin, 'ymin':ymin, 'xmax':xmax, 'ymax':ymax},

                   {...}],

 'frame_index_1': [{'id': pedestrian_name, 'xmin':xmin, 'ymin':ymin, 'xmax':xmax, 'ymax':ymax},

                   {...}],

Regarding the state of the mobile cameras in each step, this information is gathered in the "state_cam_info.json" file. The structure of each file is: 

[{'n_iteration': frame_index_0, 'pos_x': x_coordinate, 'pos_y': y_coordinate, 'pos_z': z_coordinate, 

  'orient_w': quaternios_orientation_w, 'orient_x': quaternios_orientation_x, 

  'orient_y': quaternios _orientation_y, 'orient_z': quaternios_orientation_z},

 {'n_iteration': frame_index_1, ...},

...

]

⭐ The Unreal Engine Project can be also downloaded from here .