๊ฒฝ๊ธฐ ์ฑ๋จ์ ์์ ๊ตฌ ์ฐฝ์ ๋ก 42 ํ๊ต ์ 2 ํ ํฌ๋ ธ๋ฐธ๋ฆฌ
Vision ํ
Object Detection & Depth Estimation
About darknet_detetct_and_depth_cam.sh
./darknet detector demo cfg/coco.data cfg/yolov3-tiny.cfg
[ path "yolov3-tiny.weights"]
[ -ocam ]
[ -gray4stereo]
[ -c : camera index]
[ -thres : threshold for detection probability]
[ -w : 2 X left(right) camera image width]
[ -h : left(right) camera image height]
[ -lrf : left(negative integer such as -1), right(positive integer such as 1) or full(zero) image for detection input]
[ -depth : non-zero for estismating depth, zero for no-depth estimation]
[ -both : when estmating depth, non-zero for computing both left and right disparities, zero for only left disprity.]
[ -path_agg : when semi-global block matching, 8 for 8-way path aggregation, 4 for 4-way path aggregation.]
[ -alfa : alpha (alpha = -1(default) or 0 <= alpha <= 1) value for stereoRectify function]
[ -percent_closest : for generalized median filter, 0 for the closest pixel from the camera, 100 for the furthest pixel from the camera.]
[ -class : the class name of object to measure the ditance from camera. It should one of the coco classes such as 'person']
[ -shrink 1.0 ] [ -poly : path "first_disparity_second_millimeter_3rd_polynomial_fit_first_constant_last_3rd_180920_shrink_0.5.txt"]
[ -cam_intr : path "intrinsics.yml"]
[ -cam_extr : path "extrinsics.yml"]
To run sh end2end.sh
- Set camera option and directories
- Process
1. Stereo camera Calibration and Get rectified_stereo image files
2. Save ground truth distance images ( press 'tab' keys )
3. Rectify saved images and Get disparity. Make comparison table between ground truth distance and estimated distance
4. Make real distance mapping
5. Display distance of target object