This page contains additional material for paper Cooperative Sensing and Recognition by a Swarm of Mobile Robots, A. Giusti, J. Nagi, L. Gambardella, G.A. Di Caro, submitted to IROS 2012
The working system is demonstrated, with audio commentary, in the following video. It describes a demo which has been accepted at the demo track of AAMAS 2012 (abstract).
This additional video, also attached to our IROS submission, demonstrates the system working with fingercount gestures.
This archive includes the simulator and the preprocessed test set (i.e. classification vectors obtained from images in the test set). It allows to replicate all experimental results presented in the paper and automatically recreates figures 3, 6, 7 and 8 by running the top-level MATLAB script. The process requires several hours as around 10000 different simulations are executed using different parameters.
The dataset is made available to readers in different formats.
The 22+ GB archive containing 74181 PNG images, acquired from multiple points of view, will be made available online once the paper is published.
Each image is annotated with:
The archive containing pre-segmented images is available here.
Each image is a 32x32 binary PNG, and results from color-based segmentation of original images. The largest connected component is selected and its bounding box is resized to 32x32. The resulting patch is the starting point for computing features and is saved as a PNG image.
Each image in the dataset is annotated with the following information, encoded in the filename.
For example, the image in file image.0000000073_c5_200_015.png corresponds to:
0000000073: timestamp of acquisition (in tenths of second)c5: ground truth class corresponding to finger count, 5 in this case200: distance from the hand (in centimeters)015: angle of acquisition (i.e. theta), in the following set: {270,285,300,315,330,345,000,015,030,045,060,075,090}.The 20 shape features computed from each of the images are available in this CSV file. Each line contains the features for a single image. The first column is the source segmented image filename as in the previous archive. Columns 2 to 21 contain numerical values for the feature vector.
Classification vectors (only for images in the testing set) are available in this CSV file.
The CSV contains the following columns:
The precomputed classification vectors are already included in the simulator download.
Some additional details not reported in Section II of the paper are reported in this appendix.
