act_fast.h File Reference

Finds corners in an image by actively scanning the image. More...

#include "std.h"
#include "lib/vision/image.h"

Include dependency graph for act_fast.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures
struct	agent_t

Functions
void	act_fast (struct image_t *img, uint8_t fast_threshold, uint16_t *num_corners, struct point_t **ret_corners, uint16_t n_agents, uint16_t n_time_steps, float long_step, float short_step, int min_gradient, int gradient_method, int camera_id)
	Do an ACT-FAST corner detection. More...

Detailed Description

Finds corners in an image by actively scanning the image.

This method is inspired by the work in: de Croon, G.C.H.E., and Nolfi, S. (2013, May). Act-corner: Active corner finding for optic flow determination. In Robotics and Automation (ICRA), 2013 IEEE International Conference on (pp. 4679-4684). IEEE.

The main idea of this particular implementation, called ACT-FAST, is that actively scanning the image allows to:

1. Skip uniform areas in the image. If these areas are contiguous, many non-corners will never be evaluated.
2. Follow edges. Typically, following an edge will bring an agent to a corner. These two simple rules lead to a significant lower number of corner evaluations, while still detecting a reasonable number of corners. Moreover, since the agents scanning the image start on a grid, corners will be quite well-distributed over the image. Each step of the agent starts by classifying the agent location as a corner or not with FAST.

For bigger images (e.g., 640 x 640), the computational advantage of ACT-FAST over the normal, exhaustive application of FAST becomes significant (in the order of a factor > 10).

The code here is part of the following publication: de Croon, G.C.H.E. "ACT-FAST: efficiently finding corners by actively exploring images.", in submission.

Definition in file act_fast.h.

---

Data Structure Documentation

agent_t

struct agent_t

Definition at line 28 of file act_fast.h.

Data Fields
int	active
float	preferred_dir_x
float	preferred_dir_y
float	x
float	y

Function Documentation

act_fast()

void act_fast	(	struct image_t *	img,
		uint8_t	fast_threshold,
		uint16_t *	num_corners,
		struct point_t **	ret_corners,
		uint16_t	n_agents,
		uint16_t	n_time_steps,
		float	long_step,
		float	short_step,
		int	min_gradient,
		int	gradient_method,
		int	camera_id
	)

Do an ACT-FAST corner detection.

Parameters

[in]	*img	The image to do the corner detection on
[in]	threshold	The threshold which we use for FAST9
[in]	*num_corners	reference to the amount of corners found, set by this function
[in]	**ret_corners	pointer to the array which contains the corners that were detected.
[in]	n_agents	The number of agents that will scan the image for corners
[in]	n_time_steps	The maximum number of time steps allowed for scanning
[in]	long_step	When there is not enough texture, the agent will take a long step to a next point of this length in pixels
[in]	short_step	When there is texture, the agent will follow the edge with this short step in pixels
[in]	min_gradient	The minimum gradient, in order to determine when to take a long or short step
[in]	gradient_method	0 = simple {-1, 0, 1}, 1 = Sobel {-1,0,1,-2,0,2,-1,0,1}

Definition at line 58 of file act_fast.c.

References agent_t::active, agents, fast9_detect_pixel(), FAST9_MAX_CAMERAS, image_t::h, image_gradient_pixel(), MAX_AGENTS, n_agents, n_time_steps, agent_t::preferred_dir_x, agent_t::preferred_dir_y, image_t::w, agent_t::x, point_t::x, agent_t::y, and point_t::y.

Referenced by calc_fast9_lukas_kanade().

Here is the call graph for this function:

Here is the caller graph for this function: