Technology

Let’ s spread the world’s leading edge collision prevention technology

SRIM Technology

Realize a no collision world
    using “ ultra-high speed 3D image detection technology ”

Extensively applicable high speed 3D measurement system that animal with two eyes got in the process of evolution.
Aim for practical application of 3-D image recognition technology that makes maximum use of measuring accuracy of
full field of view.
■ SRIM ( Stereo Range Imager ) Technology
  Ultrahigh-speed full-field 3D distance measuring technology

 

ITD Lab Corp. contributes to establishing “ Safe World ”,
where no crash & collision accidents, through its proprietary
“Stereo Range Imager” technology, 3D distance measurement system.

■ ADAS ( Advanced Driver Assistant System )

Stereo camera 3D distance measurement technology is a system that
automatically detects pedestrians, car, obstacles, white lines etc. and
supports safe driving.

 

 

Intelligent Stereo Camera powered by SRIM Technology
ISC 100VM is suitable for small foot print deep learning by utilizing the real time 3D distance information
Intelligent Stereo Camera

■ Why Stereo Camera?

A stereo camera that recognizes three-dimentional objects as human eyes
Detect high speed, accurate position, shape and relative speed.

A:ISC Stereo Camera B:Single Camera C:LIDAR D:Millimeter Wave Radar
3D Recognition All 3D objects Model objects only High reflectance object
Distance Accuracy 100m±2m 25m±0.08m 100m±10m 25m±0.5m ±0.1m(all range) ±0.2m(all range)
View Angle 40°~ 120° 40° 20° 10°
Detection Speed 60fps(up to 160fps) 30fps 15~20fps 10fps
Lane Detection White lane & guardrail White lane only No No
Object Separation Advantage in object
boundary detection
Detectable but inferior Slow reaction rate Low accuracy
Cost Cheaper than B/C/D
expensive as D
Equivalent and expensive Scan type control unit
is expensive
Expensive
Installation Easy due to automatic
adjustment function
Need few hours by
skilled people

■ ITD Lab ISC (Intelligent Stereo Camera) Features

High speed object recognition function, automatic adjustment function, self position recognition
function by incorporating our proprietary “SRIM Technology”, we will achieve performance that
surpasses that of other companies’ stereo cameras

Comparison with Other Stero Cameras
Intelligent Stereo Camera S / Stereo Camera Others / Stereo Camera
Self Location Recognition Capability × ×
3-D Map Generation Capability × ×
Calibration Capability ◎ GOOD △ POOR ×
Detection Speed ◎ 60fps ~ 160fps ○ 30fps △ Below 30fps
How easy to install ◎ No need to adjust × Consumer can’t install △ Consumer can’t install
Processing Speed Index ◎ 500 ○ 100 × 30~50
Cost Index ◎ 50~80 ○ 100 △ More than 200


■Advantage of Stereo Camera
1、Stereo Camera can detect objects of any distance

Comparison with LiDAR, Millimeter Wave Radar, Monocular Camera

          ●Stereo Camera → object (Mt. Fuji) on the screen, you can know the distance to that

Disparity 0 pixel → beyond 1.2 km

Parallax 60 pixels→5m

          ●Monocular Camera → Impossible to distinguish whether Mt. Fuji is genuine or a picture,
                        can not measure the distance.

Real Mt. Fuji

Fuji 5 m ahead

          ●LiDAR · Millimeter Wave Radar → can not detect Mount Fuji.

2、Stereo Camera has high accuracy of close distance

Comparison with LiDAR, Millimeter Wave Radar, Monocular Camera 2

●Distance accuracy of LiDAR and Millimeter Wave Radar is high, but distance accuracy at 10 m ahead
 is ± 4 cm even with a Stereo Camera with 300 thousand pixels.
 Equivalent to ± 5 cm (IBEO LUX company) of automotive LiDAR. It is ± 1.5 cm at 5 m and the accuracy is higher.
●Collision Avoidance Brake requires a high degree of accuracy at a distance far enough for collision time,
 and higher precision is required as approaching an object.
 As a result, LiDAR and Millimeter Wave Radar can only oversubscribe in the far side, roughly in the near direction.

3、Absolute distance can be obtained by 3D object detection
Advantage over monocular camera
●The Monocular Camera estimates the distance by finding an object matching the model of the object
 from the image. Assuming that the lower end of the object is in contact with the road surface,
 convert the lower position in the estimated horizontal line image into the distance.
●Because the Stereo Camera mechanically derives the distance from the two images,
 no models or assumptions are entered.

■Why did the UBER accident happen?

    If it is a Stereo Camera equipped with SRIM technology accident could be prevented beforehand !?

  4、the Stereo Camera has small object dependency

●Because of LiDAR and Millimeter Wave Radar capture reflected waves,
 to detect objects with low reflectance is difficult.
●Since the Camera captures the change in light and dark, even if there is an object with low
 reflectance in front of an object with high reflectance, the boundary between light
 and dark is detected, that can find an object in front.
●With a Monocular Camera, if there is no model, the detected boundary also appears
 on the pattern on the surface of the object behind, It can not be extracted as an object.
 Therefore, that can not measure the distance.
Stereo Camera knows the three-dimensional position of the boundary,
 that can find the size and shape of the object in addition to the distance.

LiDAR and Millimeter Wave Radar are hard to detect for people with black clothes.

5、The Stereo Camera has
         high lateral spatial resolution
●The relative speed in the lateral direction is important for predicting
 the collision of an object popping out from the side. For that purpose,
 it is important to detect the movement of the boundary of the object.
●Since LiDAR and Millimeter Wave Radar detect the distance to the plane,
 movement can not be detected even if the plane moves laterally.
 Since the camera is basically a sensor for detecting a pattern,
 it is possible to easily detect the boundary
 of the object and its relative speed.
●Because the distance of the object can be known from the
 Stereo Camera with respect to the Monocular Camera,
 It is possible to detect motion in all directions including
 forward and backward directions only with the Stereo Camera.

6、Processing time is short
In order to cope with sudden jumping out from the side or object tracking when sudden turning, it is necessary to
continuously grasp the surrounding situation in a short time.
The camera provides rich information of over 1 million pixels at a high speed of 60 fps or more.
In the stereo camera, it is also possible to output the distance distribution image at 160 fps by speeding up the processing of the distance calculation portion.
Monocular cameras require matching with models on the full screen, and the types of models are increasing.
Therefore, the processing time tends to increase, and the processing chip is also enlarged to suppress it.

SAD calculation circuit which is made into FPGA

        7.Stereo Camera has a large viewing angle
Millimeter wave radar is difficult to scan large and it is difficult to obtain a large viewing angle exceeding 90 degrees.
LiDAR can obtain a large viewing angle exceeding 180 degrees by using a mirror or turning the main body,
It is difficult to perform high-speed scanning like several ten frames / second because it is done mechanically.
The Monocular Camera can easily increase the viewing angle by using a fisheye lens or an omnidirectional mirror.
Even in the case of a Stereo Camera it has become possible up to about 150 degrees.


Pre-correction image

Corrected image

Super wide angle disparity
image of 150 degrees
       8.White lines and yellow lines can be detected
Lines indicating lanes such as white lines and yellow lines are very important as guidance
for automobile driving course prediction and automatic driving.
However, these can only be detected with Cameras.
     9.Ability to cope with nighttime and bad weather
Automatic driving in a bad environment where human beings can not drive, all sensors are inexperienced.
However, the current trend of the automatic operation is a simple and easy-to-operate environment in which the
low arousal level of human consciousness such as highway drive. In that case any sensor is a coarse minority.
Cameras have headlights at night and wipers with rain and snow, so that can drive like a human.

   ■ Advantages over other companies’ Stereo Cameras

  1.Being a 4K Stereo Camera

4K real-time Stereo Cameras not yet exist in the world.
That will obtain stereoscopic images in a wider view even further.
For example, if the viewing angle is set to 30 degrees, it can be measured at ± 30 m even in the distance of 500m ahead.
It is possible to run smoothly and comfortably as the veteran driver in automatic operation
if the accuracy in the distance is good.
At 30 m ahead it is comparable to Laser and Millimeter Wave Radar with ± 10 cm accuracy.
It is only ± 2.5 cm at 10 m ahead, and there are no other sensors that can ensure such accuracy.
The Bf value is 300 even at a viewing angle of 120 degrees, that can measure 10 m radius with an accuracy of ± 4 cm.

2.ITD stereo processing circuit with self adjustment function
ITD’s stereo processing circuitry has its own automatic adjustment function technology.
With this technology, the absolute distance can be adjusted by the camera itself, and adjustment at the time of attachment
and after installation is unnecessary.
It is a revolutionary technology leading to significant cost reduction not only for Stereo Cameras of other companies but also
for Laser and Millimeter Wave Radar.

Required adjustment

The relative position of the camera shifts with the lapse of time.
The cause is the twisting of the stay due to temperature
and processing strain release deflection is mostly.
 
As a result, minute rotation of the camera occurs,
distortion in the image occur.
Since its movement is minute, it is negligible, It can be corrected by
translating to the left and right and rotating around the optical axis.

3.Accurate correction calibration is high  
Stereo Camera has a weak point which is impossible to obtain the parallax of a portion where there is no image pattern.
Other Companies’ Stereo Cameras use a high-density parallax image method such as SGM (Semi Global Matching)
which achieve to obtain parallax even in the absence of patterns.
Even that brought a clean parallax image, correct calibration may not be correctly done.
So erroneous parallax diffuses and a distance distribution with low reliability is obtained.
 
For ITD Stereo Cameras, parallax-free parts are divided without parallax data, and only reliable parallax is used.
In addition, the correction calibration method is also devised, and the place where deviation of about 0.2 to 0.3 pixels appears usually
is suppressed to 0.1 pixels. Increasing the accuracy of the correction calibration increases the correct disparity density,
and the original result which is a highly accurate parallax image with sufficient density is obtained without using SGM or the like.
This is an important feature that can ensure superiority even if other companies develop 4K Cameras.

    Features of correction calibration:
        Use lattice pattern to simultaneously correct right and left images
Before correction After correction

    Five parameters out of seven parameters required for calibration are determined.
   The rest is equivalent rotation about the optical axis and up and down translation.

Parallax image by SAD only
High density disparity image by SGM

Eroding the area, the sky disappeared, the object became fat, connected,
Meaningless parallax may occur or objects may erase after being eroded by the background.

4.Fast Processing             
Besides not providing a frame buffer, the parallel pipeline processing is thoroughly advanced taking advantage of
the characteristics of the FPGA, realizing the world’s fastest stereo camera.
It also contributes greatly to SAD’s high precision and not including processing to make high-density parallax images.
The correction calibration circuit also developed its own circuit, and it has become smaller than other companies.
Furthermore, detection of three-dimensional objects and white lines is incorporated into the FPGA.
In this way, since all processes are made into FPGA, that provide the features of compact, light,
low cost and ease to upgrade the version.

                     3D object detection in real time
       Original Image                             Solids Detection
 

5.Capture the shape of the three-dimensional object as a clear parallax image
In addition to not performing interpolation which impairs accuracy, by using the luminance information
of the original image, stereoscopic images with high spatial resolution in upper, lower, right and left, clear edges,
and thus clearly understanding the shape of three-dimensional objects have been obtained.
This is also unique method of ITD.

 

For inquiries · estimate click here