cywputao / hyperion

rgb-d based people detection, tracking and prediction.

hyperion

rgb-d based people detection, tracking and prediction.

Basic Info

this is build to run on Ubuntu 18 with ROS melodic.

to run on ROS kinetic/Ubuntu 16 edits would need to be made to the build files, particularly how the eigen library is imported.

Setup

Bring in the data files.

go to hyperion/hyperion data and run sh get_data.sh. you will need to have p7zip installed (sudo apt install p7zip) to unzip the data file.

build the ROS code

everything builds with catkin_make

Run app

start ros: roscore

Detection

start publishing people tracks by running rosrun hyperion_detect hyperion_detect_node

Prediction

run a 30 second data collection and analysis with rosrun hyperion_predict hyperion_predict_node.

once the collection has run notes can be added to the file by answering the prompt. once this is done a evaluation file will be created.

Evaluation

there is a evaluation file that gived a detailed output of prediction data and it's acuracy, both on a per trial basis and on average. there is also a number of gnuplot scripts for creating plots of the data.

Python LSTM

the model for the python RNN model was created in google colab, the code can be seen here motion_prediction_lstm this model was trained on part of the atc shopping center tracks. and implemented to enhance the KF prediction.