Fast point-line feature detection for VIO(Visual-Inertial Odometry) based on gradient density. It mainly solves the time-consuming problem of line feature detection in repeated textures area. The pixel gradient changes frequently in this area which the region grow of LSD would be restrained.

We tested on the EuRoc dataset and the detection result of one frame from V1_03_difficult shows as follows.

Clone the repository

cd YOUR_WORKSPACE/
git clone https://github.com/LiuYuha0/point_line_feature_detection.git
cd point_line_feature_detection/FeatureDetect/

Opne matlab and run

main.m

NOTICE: If you want to test on your own dataset, do not forget to modify camera parameters!

In order to verify the performance of this algorithm, we tested it on whole dataset and the results of time consume show in Fig.1(V1_01_easy) and Fig.2(V1_03_difficult).