The code estimates Suspicious Action using Pictorial Human Pose Estimation and Classiﬁcation applied on on images taken from an UAV.
This work titled "Autonomous UAV for Suspicious Action Detection using Pictorial Human Pose Estimation and Classification" was published in Electronic Letter on Computer Vision and Image Analysis, Vol. 13, No. 1, pp. 18-32, 2014. Available at: http://elcvia.cvc.uab.es/article/view/582/0
The project page that includes a video and graphs can be found at: https://sites.google.com/site/amarjotsingh0720/publications/suspiciousactiondetection
Steps to run the code.
1. Open MATLAB. Change the directory to the suspicious_activity_detector_v1 folder.
2. Add all the folders and subfolders in code directory to the MATLAB path.
3. Run the following commands one after another.
compile; % This doesn't work on windows. Suitable modification must be done.
4. Change the threshold values for detection
det_pars.ubfpff_scale = 3;
det_pars.ubfpff_thresh = -0.75;
det_pars.iou_thresh = 0.9;
5. Code is setup and ready to run on a test image. Run the following commands to test.
image = imread('test_images/img_main.jpg');
[ubfdetections] = DetectStillImage2(image, 'pff_model_upperbody_final.mat', 'haarcascade_frontalface_alt2.xml', det_pars, 2);
6. If everything works fine, the result should be as shown in the attached image.
The code currently works only on linux systems. Suitable modifications can be made for windows.
If you get the error that MATLAB is out of memory, re-run the code by reducing the value of det_pars.ubfpff_scale.
Amarjot (2020). Pictorial Suspicious Action Detection (https://www.mathworks.com/matlabcentral/fileexchange/47020-pictorial-suspicious-action-detection), MATLAB Central File Exchange. Retrieved .
updated steps to run