Cause analysis of motion blur / smear caused by camera shooting moving objects

Catalog

1、 The problem background ：

2、 Problem analysis ：

3、 Motion blur / Analysis of the causes of smear ;

1） Exposure time is too long

2） Motion blur caused by relative motion

3） Excessive noise reduction intensity in time domain

4） Dual frame fusion in wide dynamic mode

5） The jelly effect ;

---

1、 The problem background ：

The application scenario of the customer camera is to identify vehicles on the highway , feedback The captured picture of the moving vehicle is blurred 、 There's a drag , The customer asked for Increase the frame rate 、 And reduce exposure time to solve this problem , The following is a summary and analysis of the causes of motion blur , And whether the customer's requirements are reasonable , Can it solve the problem ;

2、 Problem analysis ：

Take pictures of vehicles on the highway , Generally, it should be the following two cases ：

a、 The speed is too fast , The camera failed to capture the vehicle ;

b、 Can capture the vehicle , But the photos are pasted ;

If it's a problem a, Then we need to increase the frame rate to achieve , If it's a problem b, It is necessary to confirm what causes the motion blur （ Also called Montion artefact）, Too long exposure time will cause blurring 、 The generation of drag shadow , And increasing the frame rate can really reduce the exposure time , But increasing the frame rate has a great impact （mipi Rate increase , There are requirements for platform performance 、 The brightness darkens and the noise increases under low illumination ）, So you can not consider adjusting the frame rate for the time being , Confirm the specific reasons before making modifications .

3、 Motion blur / Analysis of the causes of smear ;

1） Exposure time is too long

Reasons for appearance ： When shooting a moving object , Only if the exposure time is fast enough , An object is relatively stationary relative to the camera , Otherwise, the whole running process of the moving object is cmos Imaging on , So the picture will be blurred , There's a drag ;

measures ：

a、 During debugging , First set the exposure time manual live , Reduce exposure time （ Integral time ）, See if there is any improvement , Confirm whether the exposure time is too long ;

b、 If the set maximum exposure time is reached in this scene , The maximum exposure time can be reduced , The maximum exposure time setting in the general drive （ Integral time ） yes ：VTS - frame_offset ;

c、 If the scene is not very dark , The exposure time did not reach the maximum , It is also possible to reduce the maximum exposure time to a suitable value , But at this time, due to the reduction of exposure time , The image darkens , Only by adjusting sensor Gain to compensate , The noise is bound to be serious ;

d、 The other better way is , It can be changed to high sensitivity sensor、 Large aperture lens （ It is because of insufficient exposure , So the exposure time is long ）;

e、 In addition, it is feasible to increase the frame rate , In this way, the exposure time per frame per unit time will be reduced , But the image will darken under low illumination , It will bring noise ;

2） Motion blur caused by relative motion

Reasons for appearance ：

It is quite common when we take pictures with mobile phones , His hands shook for a while , The picture will be blurred , Especially in recent cases , The camera and the subject produce relatively large relative motion , That is, the whole motion process is imaged in sensor Yes , Cause blurred imaging ;

3） Excessive noise reduction intensity in time domain

Reasons for appearance ：

commonly ISP Modules that can reduce noise in time domain , The greater the noise reduction intensity in the time domain , The more obvious the drag of moving objects ; The actual time domain noise is just a kind of noise , It is ordinary airspace noise , Because some noise is position independent , So there will be noise jitter when watching the video （ That is, the noise level at the same position between frames is different ）, Time domain noise reduction can be understood as the superposition of multiple frames before and after , So for moving objects , The drag is obvious （ The position of a moving object varies greatly from frame to frame ）;

notes ： Time domain noise reduction is not only the superposition of multiple frames , Otherwise, the jitter noise will also affect the definition , Noise reduction module , There is also a corresponding motion detection module , The moving part of the picture is denoised in time domain , Do airspace noise reduction without exercise ;

measures ：

a、 May first bypass Time domain noise reduction module , See if the shadow is gone ;

b、 It is confirmed that it is caused by time domain noise , Determine the current ambient brightness , Because the noise reduction intensity in time domain is adjusted according to the gain value （ The darker the environment , The greater the gain , The more obvious the noise , The greater the degree of noise removal ）, If the environment is still very bright , First, check whether the exposure parameters are correct ;

c、 The wrong exposure parameters are excluded , The next level of intensity time domain de-noising is called, resulting in , Then we can properly reduce the intensity of time domain noise reduction , But pay attention to the trade-off of dynamic noise ;

4） Dual frame fusion in wide dynamic mode

Reasons for appearance ： When sensor choice HDR Mode time , Two frames of long and short exposure will be output （ The corresponding is dol Of hdr,digital overlap That is, the long exposure has not yet readout End , The short exposure starts to expose ）,ISP Do double frame fusion , Output an image with a high dynamic range ;sensor During long and short exposures , There will be a time interval , When imaging a moving object , There is a time difference between the two images , So the captured pictures are different , When isp Fuse the two frames , There will be a similar phenomenon in the figure above , But this drag is not very long , And it is obvious in the highlight , So it can be used to distinguish ;

measures ： This multi frame exposure time system sensor, Due to the difference in sampling time between frames , It is impossible to avoid this kind of multi frame fusion   Montion artefact, It can be replaced by DCG （dual conversion gain） Of sensor To avoid , It's through height gain The way , Output two different exposures , There is no time difference between frames ;

5） The jelly effect ;

Reasons for appearance ： Use line by line exposure sensor Caused by , because sensor It is exposed line by line , There is a time difference between lines , When shooting fast moving objects , It will cause the imaging object to be distorted , Elongated , Pay attention to distinguish it from the smear caused by exposure time , If the exposure time is too long, there will be a smear , But it will not cause the imaged object to deform ;

measures ：

a. By increasing the frame rate 、 Reducing the time of one line can reduce the exposure time , It can alleviate the jelly effect .

b. Change global shutter （Global Shutter） Of sensor, It can completely solve the jelly effect , Global shutter sensor  It allows all pixels to start and end exposure at the same time , After exposure, read out in sequence . Because all pixels are exposed at the same time , There will be no time difference , So there will be no jelly effect when shooting fast moving objects .

notes ：global shutter and rolling shutter How to choose your camera ？

--- global shutter Shorter exposure time , But it will increase RMS Read out noise ,Rolling shutter Can achieve higher frame rate ？？, But when the exposure is improper or the object moves quickly , Partial exposure will occur (partial exposure)、 Slope graphics (skew)、 The shaking (wobble) And so on ; Applications with short exposure time （ Such as <500μs） fit global shutter, Long exposure time （ If it is greater than 500μs） when , choice rolling shutter Can have lower noise and frame rate ？？;

--- For example, it is used in vehicle 、 In machine vision Camera, For vehicles 、 Face 、 The detection and recognition of characters have high requirements , Jelly effect has a fatal impact on the detection and recognition of back-end algorithms , So we use global shutter, Security 、 Mobile phones and some consumer cameras , It's mainly for rolling shutter , They do not require high frame rate （ Security monitoring is generally 25fps/30fps）, There are certain requirements for noise under low illumination 、 The exposure time is relatively long ;

--- in addition global shutter It's usually CCD sensor, It costs more than cmos sensor high , And global shutter All pixels are exposed at the same time , A lot of data needs to be processed , There are also requirements for the performance of the platform , This aspect should also be taken into account ;

Reference resources ：

https://blog.csdn.net/passball/article/details/44301773