1 # Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
3 @page dwx_freespace_detection_sample Free-Space Detection Sample (FreeSpaceNet)
5 Drive-able collision-free space, i.e. the space that can be immediately reached without
6 collision, provides critical information for the navigation in autonomous driving.
7 This free-space example demonstrates the NVIDIA end-to-end technique of detecting the
8 collision-free space in the road scenario. The problem is modeled within a deep
9 neural network (FreeSpaceNet), with the input being a three-channel RCB image and
10 the output being a boundary across the image from the left to the right. The boundary
11 separates the obstacle from open road space. In parallel, each pixel on the
12 boundary is associated with one of the four semantic labels:
18 This FreeSpaceNet sample has been trained using RCB images
19 with moderate augmentation.
21 This sample streams a H.264 or RAW video and computes the free
22 space boundary on each frame. The sample can also be operated with cameras.
24 
26 ### Sensor details ####
28 The image datasets used to train FreeSpaceNet have been captured using a View Sekonix
29 Camera Module (SS3323) with AR0231 RCCB sensor. The camera is mounted high up at the
30 rear-view mirror position. Demo videos are captured at 2.3 MP.
32 To achieve the best free-space detection performance, NVIDIA recommends to adopt
33 a similar camera setup and align the video center vertically with the horizon before
38 The sample H264 video and camera calibration files are located at:
40 sdk/data/samples/freespace/
42 The latency of the sample FreeSpaceNet model:
43 - NVIDIA DRIVE PX<sup>™</sup> 2 with GP106: 3.98 milliseconds
44 - NVIDIA DRIVE PX 2 with iGPU: 13.82 milliseconds
46 #### Running the Sample
48 The command lines for running the sample on Linux:
50 ./sample_freespace_detection --video=<video file.h264> --rig=<calibration file.xml>
53 ./sample_freespace_detection --video=<video file.raw> --rig=<calibration file.xml>
55 The command line for running the sample on NVIDIA DRIVE PX 2 with cameras:
57 ./sample_freespace_detection --input-type=camera --camera-type=<camera_type> --csi-port=<csi_port> --rig=<calibration file.xml>
59 @note The free-space detection sample directly resizes video frames to the network input
60 resolution. Therefore, to get the best performance, it is suggested to use videos with
61 similar aspect ratio to the demo video. Or you can set the Region of Interest (ROI) to perform
62 inference on a sub-window of the full frame.
66 The free-space detection sample:
69 - Overlays ploylines for the detected free-space boundary points.
70 - Computes boundary points in car coorindate system, if a valid camera calibration file is provided.
72 The colors of the ploylines represent the types of obstacle the boundary interface with:
