Download presentation
Presentation is loading. Please wait.
Published bySabina Henry Modified over 8 years ago
1
Height Estimation from Egocentric Video- Week 1 Dr. Ali Borji Aisha Urooj Khan Jessie Finocchiaro UCF CRCV REU 2016
2
Meetings Monday- Met with Aisha and Dr. Borji Went over the goals of the project Aisha created an initial draft of the project goals and steps Tuesday- Met with Aisha Created a timeline of the project Updated project report Created Excel sheet with related papers Wednesday Implementing an SVM on images Implementing k-means clustering on image descriptors Thursday- Started data collection with Aisha Continued SVM code
3
Stages of the project 1. Data Collection 2. Estimating if the wearer is short, average, or tall 3. Estimating height as a continuous measurement 4. Improving our accuracy
4
1. Data collection Create our own dataset 10 people Place the camera at 3 locations Head Chest Waist Simulates having 30 people: 10 of each category
5
Head camera
6
Chest camera
7
Waist camera
8
2. Estimating if the wearer is short, average, or tall (Coarse-level) 3 categories of height Tall Average Short Compute Gist features Train SVM
9
3. Estimating height as a continuous measurement (Fine level) Record the height (in cm) of the camera at each recording Switch from SVM to CNN Want a continuous measurement as the output Not a category
10
4. Improving our accuracy Continue training CNN Add background movement Goal: Estimation with +/- 5 cm of actual height.
11
Timeline
12
SVM so far
13
Keypoint generation
14
Summary Video Processing Reads in video and saves every nth frame Generate data Video is data for phase 2 SVM Reads in all files in a folder, flattens them, and returns a 2D array with each image represented as a row Like MNIST Initializes, trains, and saves a SVM for the images Image Descriptors Wrote an example script using k-means clustering of the keypoints generated Implemented ORB (Oriented BRIEF) to generate local keypoints and extract image descriptors Used pyleargist to extract global image descriptors from images
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.