This repository contains the code for running the balltracking software on the Raspberry Pi GPU. Only tested on 32 bit OS, the libraries will probably not work on 64 bit. The code is built on top of the raspivid and raspistill programs, see below for how to update to newer version of those programs. Currently this uses the legacy broadcom GL driver (as opposed to Mesa + vc4-fkms-v3d), which, at the time of writing, is not available for the Raspberry Pi 4. The drivers for the Pi 4 currently do not support getting the camera data directly into an OpenGL texture, so this code will not run on a Pi 4.

There are two programs:

• The recorder, raspiballs, runs the tracking on camera input and records h264 video. Its based on raspivid.
• The video player, videotracker, runs the tracking on raw h264 video (as recorded by raspivid or raspiballs). Its based on the hello_videocube example program.

The directory structure is as follows

src/tracker
src/player
src/recorder

All important tracking code is in src/tracker, and is shared by the two programs.

This explains how the tracker works together with the web interface (https://github.com/tombana/foosball-web).

The most important files are:

• raspiballs - The program that runs the balltracking code on the GPU and detects if a goal is scored
• webproxy.py - Python script that handles communication between web interface

Other files:

• run-tracker.sh - Wrapper around raspiballs that sets correct resolution
• generate-replay.sh - Concatenates the last few replay fragments into one replay file
• player - Program that replays videos fullscreen at custom framerate. It does not do tracking.
• replay.sh - Wrapper around player

The python script webproxy.py acts as a proxy between the web interface (i.e. the javascript code) and the raspiballs program. It runs a websocket server, and when the web interface is opened, the javascript code will try to connect to the websocket server. When the web interface requests tracking, webproxy.py will run run-tracker.sh. When a goal is scored, raspiballs writes some data to a FIFO file, which is read out by webproxy.py and sent to the web interface. When the web interface requests a replay, webproxy.py will run generate-replay.sh followed by replay.sh.

This code requires cmake, make, gcc and xxd to be installed. The program xxd can be found in the vim package.

On a Raspberry Pi, the correct libraries should already be present in /opt/vc so there is no need to do anything. If the files are not present, then you can build and install them yourself as follows (from a separate directory).

git clone https://github.com/raspberrypi/userland
cd userland
./buildme

It might ask for a sudo password after building, in order to install the files into /opt/vc.

At the top of CMakeLists.txt, choose whether to do a debug or release build.

Use the buildme script to build.

The program needs to access /dev/vcsm (VideoCore Shared Memory) which by default requires root permissions. However, by placing vcsm_udev.rules in /etc/udev/rules.d, every user gets permissions to /dev/vcsm, and this way we do not need to run the tracking software using sudo.

sudo cp vcsm_udev.rules /etc/udev/rules.d/vcsm_udev.rules

To run without recording

build/raspiballs -w 1280 -h 720 -fps 40 -t 0 -g 10 --ev 5 --glwin 450,700,640,480

To record video (in fragments)

mkdir -p "/dev/shm/replay/fragments"
build/raspiballs -o /dev/shm/replay/fragments/out%04d.h264 -w 1280 -h 720 -fps 40 -t 0  -sg 100 -wr 100 -g 10 --ev 5 --glwin 450,700,640,480

See Optimizations.md for possible optimizations that might improve the performance of raspoballs.

In the userland repository, go the directory host_applications/linux/apps/raspicam. In RaspiVid.c is the required code to record video. In RaspiStill.c and RaspiTex* there is the required code to use the camera together with OpenGL. We require both recording and OpenGL, so we have to merge those sources and the easiest way is to use RaspiVid.c as a base and include some raspitex calls in there. We furthermore have to make some changes to get from RaspiTex.c to RaspiTexBalls.c.

Somewhere at the start, add an include:

#include "RaspiTex.h"

In struct RASPIVID_STATE_S { ... } add:

RASPITEX_STATE raspitex_state;

In static void default_status(RASPIVID_STATE *state) at the very end of the function add:

raspitex_set_defaults(&state->raspitex_state);
state->preview_parameters.previewWindow.width = 640;
state->preview_parameters.previewWindow.height = 360;
state->raspitex_state.width = 640;
state->raspitex_state.height = 360;

In static int parse_cmdline(int argc, const char **argv, RASPIVID_STATE *state) under parms_used = raspipreview_parse_cmdline(&state->preview_parameters, &argv[i][1], second_arg); add:

if (!parms_used)
    parms_used = raspitex_parse_cmdline(&state->raspitex_state, &argv[i][1], second_arg);

and at the very end of that same function, before return, add:

if (! state->raspitex_state.gl_win_defined)
{
   state->raspitex_state.x       = state->preview_parameters.previewWindow.x;
   state->raspitex_state.y       = state->preview_parameters.previewWindow.y;
   state->raspitex_state.width   = state->preview_parameters.previewWindow.width;
   state->raspitex_state.height  = state->preview_parameters.previewWindow.height;
}
state->raspitex_state.preview_x       = state->preview_parameters.previewWindow.x;
state->raspitex_state.preview_y       = state->preview_parameters.previewWindow.y;
state->raspitex_state.preview_width   = state->preview_parameters.previewWindow.width;
state->raspitex_state.preview_height  = state->preview_parameters.previewWindow.height;
state->raspitex_state.opacity         = state->preview_parameters.opacity;
state->raspitex_state.verbose         = state->common_settings.verbose;

In static void application_help_message(char *app_name) under raspicli_display_help(cmdline_commands, cmdline_commands_size); add:

raspitex_display_help();

In static MMAL_STATUS_T create_camera_component(RASPIVID_STATE *state) after raspicamcontrol_set_all_parameters(camera, &state->camera_parameters); and before state->camera_component = camera;, add:

status = raspitex_configure_preview_port(&state->raspitex_state, preview_port);
if (status != MMAL_SUCCESS)
{
    fprintf(stderr, "Failed to configure preview port for GL rendering");
    goto error;
}

In int main(int argc, const char **argv), after the if (state.common_settings.gps) { ... } block, right before if ((status = create_camera_component(&state)) != MMAL_SUCCESS), add:

raspitex_init(&state.raspitex_state);

and a few lines lower, comment out the block else if ((status = raspipreview_create(&state.preview_parameters)) != MMAL_SUCCESS) { ... }.

//else if ((status = raspipreview_create(&state.preview_parameters)) != MMAL_SUCCESS)
//{
//   vcos_log_error("%s: Failed to create preview component", __func__);
//   destroy_camera_component(&state);
//   exit_code = EX_SOFTWARE;
//}

A few lines lower, comment out the line:

//preview_input_port  = state.preview_parameters.preview_component->input[0];

A few lines lower there is if (state.preview_parameters.wantPreview ) { ... } else { ... }. Comment out the whole if-else block, both the if and else paths. Both these blocks correspond to the if (! state.useGL) { ... } block that is in RaspiStill.c at that same location. The if-else block will connect the preview_port but now we used the raspitex_configure_preview_port instead, and therefore we don't need it.

//if (state.preview_parameters.wantPreview )
//{
//    if (state.raw_output)
//    ...
//    ...
//}
//else
//{
//    if (state.raw_output)
//    ...
//    ...
//}

In the same function, quite a bit lower, there is status = mmal_port_enable(encoder_output_port, encoder_buffer_callback);. After that, but before if (state.demoMode), add:

if (raspitex_start(&state.raspitex_state) != 0)
    goto error;

In the same function, near the end, there is fprintf(stderr, "Closing down\n");. After that, before check_disable_port(camera_still_port);, add:

  raspitex_stop(&state.raspitex_state);
  raspitex_destroy(&state.raspitex_state);

The main edits are:

1. Ignore requested scene and always use the balltrack scene
2. Limit drawing framerate to source framerate: in preview_process_returned_bufs only call raspitex_draw when a new buffer is available from the camera source.

Remove all the gl_scenes/* includes and replace them by only the balltrack.h file:

//#include "gl_scenes/mirror.h"
//#include "gl_scenes/sobel.h"
//#include "gl_scenes/square.h"
//#include "gl_scenes/teapot.h"
//#include "gl_scenes/vcsm_square.h"
//#include "gl_scenes/yuv.h"
#include "gl_scenes/balltrack.h"

In static int preview_process_returned_bufs(RASPITEX_STATE* state) at the end of the function, remove the raspitex_draw call.

//if (! new_frame)
//    rc = raspitex_draw(state, NULL);

In int raspitex_init(RASPITEX_STATE *state) comment out the switch (state->scene_id) and replace it with a call to balltrack_open:

//switch (state->scene_id)
//{
//case RASPITEX_SCENE_SQUARE:
//   rc = square_open(state);
//   break;
//case RASPITEX_SCENE_MIRROR:
//   rc = mirror_open(state);
//   break;
//case RASPITEX_SCENE_TEAPOT:
//   rc = teapot_open(state);
//   break;
//case RASPITEX_SCENE_YUV:
//   rc = yuv_open(state);
//   break;
//case RASPITEX_SCENE_SOBEL:
//   rc = sobel_open(state);
//   break;
//case RASPITEX_SCENE_VCSM_SQUARE:
//   rc = vcsm_square_open(state);
//   break;
//default:
//   rc = -1;
//   break;
//}
rc = balltrack_open(state);

Change this include: (RaspiTex.c will work without the include; RaspiTexUtil.c needs it)

//#include "interface/khronos/include/EGL/eglext_brcm.h"
#include "EGL/eglext_brcm.h"