Ask Your Question

Revision history [back]

If your pedestrians are quite different from pedestrians used to train the SVM, you may have to train your own SVM with your data. To do so, record many sequences with your camera, manually extract bounding box around pedetrian in each part of your image (they are positif samples), extract bounding box around non pedestrian (they are negative samples). Compute HOG on your samples and train your SVM. OpenCV has all the classes to help you during this process.