chychiang/tmod

Tracking Moving Objects of Different Classes (TMOD)

Tracking Moving Objects of Different Classes (TMOD)

Project Description

This project aims to solve the problem of tracking moving objects of different classes in a scene. Data will be in the form of a video captured by a stationary camera. We will label object classes and position. This project is particularly interesting and useful to solve because accurate tracking of objects in a scene is critical for many downstream applications, such as self-driving, path planning, and scene understanding.

tmod's Issues

object detection model

• somehow predict the bounding box accurately
• talk to instructors to see if we can use alternative metrics, such as IoU

object in scene model

a model that predicts whether or not the bottle is in the scene

scene classification

choose 2 classifier to predict what "scene" is given
i.e. living room, dining room, ...

feature extraction check point

What to do

To do this part, collect a dataset and and then extract features from this dataset using at least 3 substantially different feature extraction methods discussed in class. For example, LDA and Kernel LDA are considered similar to each other, PCA, Kernel PCA and LSA are considered similar, k-Means and Gaussian clustering are considered similar. Of the 3 methods, at least one of them should be used to show us a visualization of the dataset, and at least one of them should be used for subsequent regression or classification. For image/video/audio/text applications, at least one method must be a Manual feature extraction method. You can use a feature extraction method you have read about in a publication, if that is more suited to your application. If you spend significant effort in collecting data (such as making measurements yourself, say using a cell phone, instead of simply downloading data), you only need show us one feature extraction method.

What to submit

• A small sample of your dataset. This can be a zip file or a text, csv or excel file.
• Any software code you have developed for Feature extraction. This can be a zip file or a Matlab file or Python notebook.
• A picture showing visualization of features extracted from your dataset, as specified below. This must be a single image, pdf or word file. Do not upload a zip file for this part.

Todos

• collect data: record videos of objects in a scene
• develop code for one manual feature extraction method
• develop code for one automatic feature extraction method
• visualize extracted features for presentation (use the same slides for thie entire project - submit just the feature extraction section)