Author Archives: Yves Daoud

Led by Ph.D. Alex Spitzer, this work aimed at pushing the limits of an autonomous drone flying at high speeds using vision-based state estimation. The dynamics and disturbances experienced by the quadrotor were learned on-the-fly achieving low tracking errors and improved agility. The drone hit a max. speed of 17m/s (38mph) and more than 1g …

Continue reading “High Speed Vision-Based Quadrotor Flight”

This work develops a methodology for collaborative human-robot exploration that leverages implicit coordination. Most autonomous single- and multi-robot exploration systems require a remote operator to provide explicit guidance to the robotic team. Few works consider how to embed the human partner alongside robots to provide guidance in the field. A remaining challenge for collaborative human-robot …

Continue reading “Collaborative Human-Robot Exploration via Implicit Coordination”

Using Iterative Closest Point algorithm (ICP) and point-based fusion on raw RGB-D images of a scene, a dense 3D map is reconstructed.

I built a 3D uncalibrated photometric stereo technique as part of the computer vision class at CMU. Photometric stereo is a computational imaging method to determine the shape of an object from its appearance under a set of lighting directions. The object’s lighting directions and shape are simultaneously optimized for given a set of input …

Continue reading “3D Photometric Stereo”

I built a text recognition algorithm using a neural network as part of the computer vision class at CMU. The network is a custom-built multi-layered perceptron with Xavier weights initialization, softmax, ReLU, and sigmoid activation functions.

I built a 3D scene reconstruction pipeline as part of the computer vision class at CMU. Given a pair of stereo images and the intrinsic and extrinsic matrices of the cameras, a scene is reconstructed in as 3D sparse features using triangulation, epipolar geometry, RANSAC, and finally bundle adjustment.

I built a Lucas-Kanade object and motion tracker as part of the computer vision class at CMU. To achieve this, a pixel-wise cost function is minimized under the brightness constancy assumption resulting in a patch tracker.

I built an AR demo as part of the computer vision class at CMU. Using feature extraction, descriptor matching, planar homographies, and 3D transforms, we can replace the Computer Vision book cover with a scene from Kung Fu Panda!

I built a scene classifier as part of the computer vision class at CMU. To achieve this, images of scenes are fed into the algorithm which then extracts features, puts them into clusters, and builds a dictionary 0f visual words. The features are then clustered within each category to form a bag of visual words …

Continue reading “Spatial Pyramid Matching for Scene Classification”

This quadrotor aerial system has dimensions of 0.6m x 0.3m x 0.3m and a mass of 2.5kg. The vehicle is equipped with a downward-facing mvBlueFox-MLC200wC color global shutter camera and Sunex DSL219D-650-F2.0 lens to estimate state. A forward-facing Intel Realsense D455 is used to estimate depth. Cree Xlamp XM-L2 High Power LEDs (Cool White 6500K) …

Continue reading “Build of an Autonomous Aerial Robot”