Method

Multiview Random Forest of Local Experts Combining RGB and LIDAR data for Pedestrian Detection [la] [MV-RGBD-RF]


Submitted on 11 Jun. 2015 11:12 by
Alejandro Gonzalez (Computer Vision Center (CVC/UAB))

Running time:4 s
Environment:4 cores @ 2.5 Ghz (C/C++)

Method Description:
Multiview Random Forest of Local Experts Combining RGB and LIDAR data
Parameters:
Random Forest of 100 trees (7 levels) with liblinear as weak classifier and HOGLBP as features, in RGB images and LIDAR data. Using 2 view frontal and lateral.
Latex Bibtex:
@InProceedings{Gonzalez2016TCYB,
author="Alejandro Gonzalez
and David Vazquez
and Antonio Lopez
and Jaume Amores",
title="On-Board Object Detection: Multicue, Multimodal, and Multiview Random Forest of Local Experts.",
booktitle="IEEE Trans. on Cybernetics",
year="2016",
}
@InProceedings{AGonzalez2015,
author="Alejandro Gonzalez
and Gabriel Villalonga
and Jiaolong Xu
and David Vazquez
and Jaume Amores
and Antonio Lopez",
title="Multiview Random Forest of Local Experts Combining RGB and LIDAR data for Pedestrian Detection",
booktitle="IEEE Intelligent Vehicles Symposium (IV)",
year="2015",
}

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Car (Detection) 77.89 % 70.70 % 57.41 %
Pedestrian (Detection) 72.99 % 56.18 % 49.72 %
Cyclist (Detection) 51.10 % 40.94 % 34.83 %
This table as LaTeX

2D object detection results.
This figure as: png eps pdf txt gnuplot



2D object detection results.
This figure as: png eps pdf txt gnuplot



2D object detection results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker