Camera and Object Detection
What is the tutorial about?
In this tutorial you learn how to use a camera and HDMI with an FPGA. You learn how to apply filters to the received images and finally create an object detection.
Why an FPGA?
While a processor has to:
- Save the image from the camera in the RAM
- Read the first pixel from the RAM
- Apply the first filter
- Write the new value back in the RAM
- Repeat that for all 300k pixel
- Repeat everything for each filter
An FPGA does everything at once. The first pixel from the camera is received and goes to the first filter. The last pixel goes to the second filter and so on. You have multiple components that all work independently and don't reduce the quality or frame rate. Also the delay between receiving and the last filter is minimal.
What you need
Important: This are only links for the German products.
*This links are Affiliate Links. By purchasing with them you support us and our work, because we get a part of the revenue as commission. You still pay the same price.
Structure of the camera libraries
Program own object detection
We have a really big, but well structured example for image detection, where you can see how all components are connected.
Click here to open the example on GitHub.
In the example, yellow, black, blue and white blobs are found. If there is a bigger yellow blob on the bottom and a smaller black or yellow blob on the top, these blobs are seen as one object and a cross is displayed at the position. The same for blue and white blobs. Finally Yellow/Black and Blue/White cones should be found in the image.
You have a lot of possibilities to change the HDMI output and optimize the object detection. Here are the first lines:
Camera_CLK_Lane : IN STD_LOGIC;
Camera_Data_Lane : IN STD_LOGIC_VECTOR (1 downto 0);
Camera_Enable : OUT STD_LOGIC;
Camera_SCL : INOUT STD_LOGIC;
Camera_SDA : INOUT STD_LOGIC;
oHDMI_TX : OUT STD_LOGIC_VECTOR(2 downto 0);
oHDMI_CLK : OUT STD_LOGIC;
iHDMI_HPD : IN STD_LOGIC;
CONSTANT CLK_Frequency : NATURAL := 48000000;
CONSTANT Row_Buf : BOOLEAN := true; --Uses more RAM, but helps with less noise
--Color Correction and Threshold
CONSTANT Debug_Mode : NATURAL := 0; --To change Threshold and Color Correction parameters with ISSP
--0 = No debug
--1 = Color Correction (ISSP1 = G* ISSP2 = G/ ISSP3 = B* ISSP4 = B/ -> R* = 1 R/ = 1)
--2 = Threshold Yellow (ISSP1 = H- ISSP2 = H+ ISSP3 = S- ISSP4 = V- -> S+ = 255 V+ = 255)
--3 = Threshold Black (ISSP1 = H- ISSP2 = H+ ISSP3 = S+ ISSP4 = V+ -> S- = 0 V- = 0)
--4 = Threshold Blue (ISSP1 = H- ISSP2 = H+ ISSP3 = S- ISSP4 = V- -> S+ = 255 V+ = 255)
--5 = Threshold White (ISSP1 = H- ISSP2 = H+ ISSP3 = S+ ISSP4 = V- -> S- = 0 V+ = 255)
CONSTANT CC_G_Mult : NATURAL := 3; --(Green * ...) / ...
CONSTANT CC_G_Div : NATURAL := 5;
CONSTANT CC_B_Mult : NATURAL := 1; --(Blue * ...) / ...
CONSTANT CC_B_Div : NATURAL := 1;
CONSTANT Ye_H_Min : NATURAL := 170; --Hue mininmum value for yellow
CONSTANT Ye_H_Max : NATURAL := 45; --Hue maximum value for yellow
CONSTANT Ye_S_Min : NATURAL := 100; --Saturation minimum value for yellow
CONSTANT Ye_V_Min : NATURAL := 60; --Brightness minimum value for yellow
CONSTANT Bk_H_Min : NATURAL := 0; --Hue mininmum value for black
CONSTANT Bk_H_Max : NATURAL := 255; --Hue maximum value for black
CONSTANT Bk_S_Max : NATURAL := 70; --Saturation maximum value for black
CONSTANT Bk_V_Max : NATURAL := 60; --Brightness maximum value for black
CONSTANT Bl_H_Min : NATURAL := 105; --Hue mininmum value for blue
CONSTANT Bl_H_Max : NATURAL := 150; --Hue maximum value for blue
CONSTANT Bl_S_Min : NATURAL := 120; --Saturation minimum value for blue
CONSTANT Bl_V_Min : NATURAL := 50; --Brightness minimum value for blue
CONSTANT Wh_H_Min : NATURAL := 0; --Hue mininmum value for white
CONSTANT Wh_H_Max : NATURAL := 255; --Hue maximum value for white
CONSTANT Wh_S_Max : NATURAL := 120; --Saturation maximum value for white
CONSTANT Wh_V_Min : NATURAL := 200; --Brightness minimum value for white
CONSTANT Enable_Compression : BOOLEAN := true; --Uses more RAM, decreases noise
CONSTANT Max_Area : NATURAL := 4; --Bigger Area = less noise but higher RAM usage
CONSTANT Min_Area : NATURAL := 2;
CONSTANT Min_Pixel_Num : NATURAL := ((Max_Area**2)/2)*1; --min 50% correct color
CONSTANT Blob_Min_H : NATURAL := 3; --< = Detect smaller blobs
CONSTANT Blob_Min_W : NATURAL := 3;
CONSTANT Blob_Max_H : NATURAL := 20; --> = Detect bigger blobs
CONSTANT Blob_Max_W : NATURAL := 40;
CONSTANT Cone_31_Dist_Mult : NATURAL := 3; --Bigger = detect more cones but less accurate
CONSTANT Cone_32_Dist_Mult : NATURAL := 2;
--Capture and video output parameters
--|Nr.| Output | Color type | Recommended color depth | Recommended compression |
--| | | (Force_Mono = false) | -> Capture_Color_Depth | -> Capture_Compression |
--| 0 | Camera | RGB | 4 | 4-5 |
--| 1 | Color Filter | RGB | 4 | 4-5 |
--| 2 | HSV | Mono (only Hue) | 8 | 4-5 |
--| 3 | Threshold with Color_Select color | BW | 1 | 1-2 |
--| 4 | Output 3 after Area Compression | BW | 1 | 1-2 |
--| 5 | Blobs (ISSP) | RGB | 1 | 2 |
--| 6 | Output 7 with marked cones (ISSP) | RGB | 1 | 2 |
--| 7 | Threshold with all colors | RGB | 1 | 2 |
CONSTANT Capture_Output : NATURAL := 6;
CONSTANT Force_Mono : BOOLEAN := false; --true forces the image to monochrome
CONSTANT Capture_Color_Depth : NATURAL := 1; --How many bits for each color
CONSTANT Capture_Compression : NATURAL := 2; --Higher value = less RAM but also less resolution
CONSTANT Full_Image : BOOLEAN := true; --true -> full image with less resolution | false -> part of image with full resolution
SIGNAL Color_Select : NATURAL range 0 to 3 := 2; --0 = Yellow 1 = Black 2 = Blue 3 = White
You can set Capture_Output to 5 so you only output one color. By default the selected color is Blue, so you can change the parameters for blue blobs (
Bl_H_Max, ...) to try your own colors.
Important: Check the description of the CSI_Camera component. Here you can see what you have to do so the differential IOs for camera and HDMI work properly.
Make sure you set Camera_Enable to '1'
It is pretty difficult to handle this many components, but if you take the example and adapt it to your needs, the object detection is easy like with a processor but at a way higher performance. This Project also has a lot of potential to be expanded and be used for Robots and other smart, high speed products.
We hope you enjoyed the tutorial and feel free to check out