Human Detection using depth Zach Robertson
Papers Read Object Detection with Discriminatively Trained Part-Based Models A little on HOG Mean Shift
Problems Alignment of RGB and Depth Images Segmentation Human Detection
Initial Tests Removing background using masks based on the depth data Dilating the masks Applying Petro’s algorithm on the result
Further Tests Using Edison code to segment depth data then apply mask At each depth level Petro’s algorithm was applied Allowed for the threshold to be increased significantly, from -0.3 to -1.1
Petro’s Algorithm with no depth, low and high threshold
Third Test Align depth and RGB images then apply Petro’s algorithms
Petro’s Algorithm with no depth, low and high threshold
Future Work Apply head and shoulder detector Apply head detector Training Petro’s algorithms with depth images