CN105979120A - Defogging system and video defogging method based on distributed calculation - Google Patents

Abstract

The invention discloses a defogging system and a video defogging method based on distributed calculation. The defogging system comprises an input node, a transmissivity estimation module, an atmosphere illumination estimation module, a fog-free image generation module and an output node which are deployed on a distributed calculation architecture; and each module has a plurality of computers which are in charge of execution in a distributed architecture mode. The video defogging method comprises steps: an inputted fogged video is split into frame streams; the transmissivity estimation module simultaneously calculates the transmissivity of multiple frames; the atmosphere illumination estimation module calculates atmosphere illuminance of a corresponding frame; the fog-free image generation module generates a fog-free image corresponding to a fogged frame; and the output node outputs the fog-free image frames according to the correct frame sequence. The defogging system and the video defogging method have higher processing speed and have the real-time processing capability.

Description

Video mist elimination system based on Distributed Calculation and video defogging method

Technical field

The present invention relates to computer vision field, be more particularly related to a kind of video mist elimination based on Distributed Calculation System and video defogging method.

Background technology

The video data that haze weather causes video equipment to be absorbed is smudgy, seriously reduces the quality of data.Mesh Mark tracking system, target identification system etc. are highly dependent on the application general of video data definition therefore cannot normal operation. Video mist elimination can have mist video as input using low definition, after a series of process, exports the video data without mist.Cause This, analysis and the process of video data are extremely important by video mist elimination.

The method of video mist elimination is all just for removing fog effect at present, and instead of focusing on its performance, thus generally there is effect Rate bottleneck, it is impossible to reach the requirement processed in real time.But, most video data analyzes system, clear at video data While having particular/special requirement on degree, often also need to the ability possessing process in real time.Video mist elimination processing speed the most slowly, Will be unable to meet the performance indications of each big processing system for video, cause its practicality the highest, promotion and application also receive limitation. On the other hand, due to some the intrinsic basic calculation step involved by major part video mist elimination algorithm, itself is calculating Time there is higher time complexity, attempt improving whole efficiency from video mist elimination algorithm itself and deposit the most to a certain extent In difficulty.The treatment effeciency improving video defogging method the most further is a key and the problem having realistic meaning.

Summary of the invention

It is an object of the invention to improve further the processing speed of video mist elimination, to reach wanting of process in real time Ask.

For achieving the above object, the invention discloses video mist elimination system based on Distributed Calculation, use Apache Storm realizes framework as Distributed Calculation, and including being deployed in input node on distributed computing architecture, absorbance estimates mould Block, atmosphere light module by estimate, without mist image generation module, output node, each module divides 5 layers of deployment: the 1st layer is input joint Point, reads stream of video frames；2nd layer is absorbance estimation module, is made up of multiple nodes, and each node is deployed with an independence Absorbance estimation unit, in order to estimate the absorbance corresponding to single image frame, and result is exported to the 3rd layer；3rd layer is Atmosphere light module by estimate, is made up of multiple nodes, and each node is deployed with an independent atmosphere light unit by estimate, uses To receive the absorbance of the 2nd layer and to estimate the atmosphere light illumination of correspondence, result is exported to the 4th layer；4th layer is raw without mist image Becoming module, be made up of multiple nodes, each node is deployed with an independent image generation unit without mist, in order to receive the 3rd The atmosphere light illumination of layer output, generates without mist picture frame simultaneously, and exports the 5th layer；5th layer is output node, is used for exporting Video flowing.

Further, the input node of described the 1st layer realizes with the Spout in Apache Storm, the 2nd～5 layer All nodes realize with the Blot in Apache Storm.

Wherein, described output node will export video without mist picture frame according to after the correct order sequence of video sequence Stream.

Preferably, described input node, absorbance estimation module, atmosphere light module by estimate, mould is generated without mist image Block, output node, averagely it is fitted on some computers operation.

The video defogging method based on Distributed Calculation that the invention also discloses, comprises the following steps:

S1. by having mist video using the form of frame stream as input, with the speed of n frame per second, by input node input figure As frame I_t1,I_t2,…,I_tnIn Redis Buffer Pool；

S2. input node constantly reads from Redis Buffer Pool the picture frame of mist, and exports absorbance estimation module In；

S3. absorbance estimation module estimates absorbance T of n two field picture simultaneously_t1,T_t2,…,T_tn, and result is input to greatly In gas illuminance estimation module；

S4. atmosphere light illumination estimation module is according to picture frame I_t1,I_t2,…,I_tnAbsorbance T_t1,T_t2,..,T_tn, simultaneously Estimate atmosphere light illumination A of its correspondence_t1,A_t2,…,A_tn, and result is exported without in mist image generation module；

S5. without mist image generation module according to I_t1,I_t2,…,I_tnAnd T_t1,T_t2,..,T_tnAnd A_t1,A_t2,…,A_tn, raw Become without mist frame J_t1,J_t2,…,J_tn；

S6. to without mist frame J_t1,J_t2,…,J_tnIt is ranked up according to correct sequence order, obtains sorted frame without mist J_t,J_t+1,…,J_t+n-1, wherein J_tRepresent J_t1,J_t2,…,J_tnThe frame without mist that middle sequence order is the most front；

S7. by the sorted J of frame without mist_t,J_t+1,…,J_t+n-1Output is to Redis Buffer Pool；

S8. constantly read without mist frame from Redis Buffer Pool, recombine video flowing and exported by output node.

Wherein, the absorbance estimation module in described step S3, it is made up of multiple absorbance estimation units, each absorbance The implementation of estimation unit is the most identical, and it is implemented as follows:

A given frame has mist image I, and absorbance estimation unit calculates the absorbance of its correspondence:

$t\left(x\right)=e^{-\mathrm{\β}\underset{y=\mathrm{\Ω}\left(x\right)}{\min }\[{\mathrm{\α}}_0+{\mathrm{\α}}_1{I}^{val}\left(y\right)+{\mathrm{\α}}_2{I}^{sat}\left(y\right)\]},$

Wherein, t is the absorbance corresponding to I, and Ω (x) is the localized mass centered by x of a size of 15 15, and β is big Gas scattering coefficient, β=1.0, I^valAnd I^satIt is the brightness under hsv color space and saturation respectively, α₀、α₁And α₂For linear system Number, α₀=0.1893, α₁=1.0267, α₂=-1.2966.

Wherein, atmosphere light illumination estimation module in step S4, it is made up of multiple atmosphere light illumination estimation units, each estimates The implementation of meter unit is the most identical, and it is implemented as follows:

$A=I\left(x\right),x\∈\left\{x\right|\∀y:t\left(y\right)>t\left(x\right)\}$

Wherein, A represents the atmosphere light illumination corresponding to picture frame I.

Wherein, without mist image generation module in step S5, by multiple image generation units without mist, each unit all uses same The method of sample is restored without mist image, and concrete restored method is as follows:

According to having mist image I, absorbance t, atmosphere light illumination A, in conjunction with atmospherical scattering model, can be calculated without mist frame by following formula J, thus restore without mist image:

$J\left(x\right)=\frac{I\left(x\right)-A}{max\left(t\right(x),0.1)}+A.$

Compared with prior art, the method have the advantages that

1) combine distributed computing technology, mist video can be had multiframe parallel to process, solve the place of video mist elimination Reason efficiency bottle neck.Process time to video is greatly shortened, has possessed the ability of process in real time.

2) there is higher extensibility, by constantly extending the quantity of computer, can further improve Video processing Efficiency, and there is no ceiling restriction, its efficiency depends primarily on average behavior and the quantity thereof of computer.

Accompanying drawing explanation

Fig. 1 be the present invention method in involved each module deployment form in distributed structure/architecture.

Fig. 2 is that the method for the present invention performs step schematic diagram.

Fig. 3 is the treatment effeciency cartogram of the inventive method.

Detailed description of the invention

The present invention will be further described below in conjunction with the accompanying drawings, but embodiments of the present invention are not limited to this.

As it is shown in figure 1, the invention discloses video mist elimination system based on Distributed Calculation, Apache Storm is used to make Framework is realized, including being deployed in input node on distributed computing architecture, absorbance estimation module, air for Distributed Calculation Illumination estimation module, without mist image generation module, output node.

In Fig. 1, each circle represents a node in Distributed Architecture, represents a separate threads accordingly.Each module is divided 5 layers of deployment, detailed description specific as follows.

1st layer, as input node, is read stream of video frames, and this node realizes with the Spout in Apache Storm, remaining All nodes all realize with the Bolt in Apache Storm.

2nd layer is absorbance estimation module, is made up of multiple nodes, and each node is all deployed with an independent absorbance Estimation unit, in order to estimate the absorbance corresponding to single image frame, and exports the 3rd layer by result, between different nodes Estimation unit separate.

3rd layer is atmosphere light illumination estimation module, similar with the 2nd layer, is made up of multiple nodes, each node deployment one Independent atmosphere light illumination estimation unit, internodal unit is the most separate, each node receive at one's leisure the 2nd layer defeated The absorbance gone out, estimates the atmosphere light illumination of correspondence, and exports the 4th layer.

4th layer is without mist image generation module, is decomposed into multiple image generation unit without mist, each unit portion independently Administration, on a single node, is responsible for receiving the atmosphere light illumination of the 3rd layer of output, generates without mist image simultaneously, and output is to the 5th Layer.

5th layer is single output node, for the picture frame without mist of output according to video sequence correct the most again Sequence, and outputting video streams.

Given m platform physical computer, all separate threads (include the estimation in each input node, absorbance estimation module Estimation unit in unit, atmosphere light illumination estimation module, without the signal generating unit in mist image generation module and output node) Operation will be evenly distributed on this m platform computer.During as it is shown on figure 3, separate threads is assigned on 3 physical computers run, When same resolution, its frame per second is noticeably greater than 1 computer, greatly improves treatment effeciency, can meet process in real time Requirement.

Based on said system, the invention also discloses the processing method of video mist elimination, it performs step as shown in Figure 2, Specifically include following steps:

S1. by having mist video using the form of frame stream as input, with the speed of n frame per second, input picture frame I_t1, I_t2,…,I_tnIn Redis Buffer Pool.

S2. input node reads from Redis Buffer Pool the picture frame of mist, and exports absorbance estimation module；

S3. absorbance estimation module is made up of multiple absorbance estimation units, and each absorbance estimation unit is all in accordance with having Mist image I calculates the absorbance of its each picture frame, and computing formula is as shown in Equation 1.

Wherein, t is the absorbance corresponding to I, and Ω (x) is the localized mass centered by x of a size of 15 15, and β is big Gas scattering coefficient, β=1.0, I^valAnd I^satIt is the brightness under hsv color space and saturation respectively, α₀、α₁And α₂For linear system Number, α₀=0.1893, α₁=1.0267, α₂=-1.2966.

Absorbance T of n two field picture is obtained according to formula 1_t1,T_t2,…,T_tn, and result is input to the estimation of atmosphere light illumination In module.

S4. atmosphere light illumination estimation module is made up of, according to picture frame I multiple atmosphere light illumination estimation units_t1, I_t2,…,I_tnAbsorbance T_t1,T_t2,..,T_tn, it is corresponding that each atmosphere light illumination estimation unit calculates picture frame by formula 2 Atmosphere light illumination A_t1,A_t2,…,A_tn.And result is exported without in mist image generation module.

Wherein, A represents the atmosphere light illumination corresponding to picture frame I.

S5. it is made up of multiple image generation units without mist, according to I without mist image generation module_t1,I_t2,…,I_tnAnd T_t1, T_t2,..,T_tnAnd A_t1,A_t2,…,A_tn, each image generation unit without mist calculates the J of frame without mist generated according to formula 3_t1, J_t2,…,J_tn。

S6. output node is to without mist frame J_t1,J_t2,…,J_tnIt is ranked up according to correct sequence order, obtains sequencing sequence The J of frame without mist_t,J_t+1,…,J_t+n-1, wherein J_tRepresent J_t1,J_t2,…,J_tnThe frame without mist that middle sequence order is the most front；

S7. output node is further by the sorted J of frame without mist_t,J_t+1,…,J_t+n-1Output is to Redis Buffer Pool；

S8. constantly read without mist frame from Redis Buffer Pool, recombine video flowing and exported by output node.

To sum up, the mist video that has of input will be split framing stream by the inventive method, and the form with every number of seconds frame is incoming Penetrating the computing module of rate, this module calculates the absorbance of number frame simultaneously, and result exports the estimation module of atmosphere light illumination, Atmosphere light illumination estimation module calculates the atmosphere light illumination of corresponding frame simultaneously, and result is exported the generation mould without mist image Block, has the image without mist corresponding to mist frame without mist image generation module, and output node is defeated according to the correct sequence order of frame Go out.System and method of the present invention can be used for restoring distant view and close shot, and after recovery, the profile of distant view is obvious, and color is the most true to nature, closely The pictograph of scape is high-visible, and each module of the present invention all has the mode of multiple stage computer framework to be in a distributed manner responsible for execution, Pipeline system operates, and has higher processing speed, possesses processing capability in real time, is suitable to popularization and application.

The embodiment of invention described above, is not intended that limiting the scope of the present invention.Any at this Amendment, equivalent and improvement etc. done within bright spiritual principles, should be included in the claim protection of the present invention Within the scope of.

Claims (8)

1. video mist elimination system based on Distributed Calculation, uses Apache Storm to realize framework as Distributed Calculation, It is characterized in that: include being deployed in input node on distributed computing architecture, absorbance estimation module, atmosphere light mould by estimate Block, without mist image generation module, output node, each module divides 5 layers of deployment:

1st layer is input node, reads stream of video frames；

2nd layer is absorbance estimation module, is made up of multiple nodes, and each node is deployed with an independent absorbance and estimates Unit, in order to estimate the absorbance corresponding to single image frame, and exports the 3rd layer by result；

3rd layer is atmosphere light module by estimate, is made up of multiple nodes, and each node is deployed with an independent atmosphere light and shines Estimation unit, in order to receive the absorbance of the 2nd layer and to estimate the atmosphere light illumination of correspondence, exports the 4th layer by result；

4th layer is without mist image generation module, is made up of multiple nodes, and each node is deployed with an independent image without mist Signal generating unit, in order to receive the atmosphere light illumination of the 3rd layer of output, generates without mist picture frame simultaneously, and exports the 5th layer；

5th layer is output node, for outputting video streams.

Video mist elimination system based on Distributed Calculation the most according to claim 1, it is characterised in that: described the 1st layer Input node with in Apache Storm Spout realize, all nodes of the 2nd～5 layer are with in Apache Storm Blot realizes.

Video mist elimination system based on Distributed Calculation the most according to claim 1 and 2, it is characterised in that: described is defeated Egress will be without mist picture frame according to outputting video streams after the correct sequentially sequence of video sequence.

Video mist elimination system based on Distributed Calculation the most according to claim 3, it is characterised in that: described input joint Point, absorbance estimation module, atmosphere light module by estimate, without mist image generation module, output node, be averagely fitted on some meters Run on calculation machine.

5. the video defogging method of the video mist elimination system based on Distributed Calculation described in any one of Claims 1 to 4, it is special Levy and be, comprise the following steps:

S1. by having mist video using the form of frame stream as input, with the speed of n frame per second, input picture frame I_t1,I_t2,…,I_tn In Redis Buffer Pool；

S2. input node constantly reads from Redis Buffer Pool the picture frame of mist, and exports in absorbance estimation module；

S3. absorbance estimation module estimates absorbance T of n two field picture simultaneously_t1,T_t2,…,T_tn, and result is input to atmosphere light In illumination estimation module；

S4. atmosphere light illumination estimation module is according to picture frame I_t1,I_t2,…,I_tnAbsorbance T_t1,T_t2,..,T_tn, estimate simultaneously Go out atmosphere light illumination A of its correspondence_t1,A_t2,…,A_tn, and result is exported without in mist image generation module；

S5. without mist image generation module according to I_t1,I_t2,…,I_tnAnd T_t1,T_t2,..,T_tnAnd A_t1,A_t2,…,A_tn, generate nothing Mist frame J_t1,J_t2,…,J_tn；

S6. to without mist frame J_t1,J_t2,…,J_tnIt is ranked up according to correct sequence order, obtains the sorted J of frame without mist_t, J_t+1,…,J_t+n-1, wherein J_tRepresent J_t1,J_t2,…,J_tnThe frame without mist that middle sequence order is the most front；

S7. by the sorted J of frame without mist_t,J_t+1,…,J_t+n-1Output is to Redis Buffer Pool；

S8. constantly read without mist frame from Redis Buffer Pool, recombine video flowing and exported by output node.

Video defogging method based on Distributed Calculation the most according to claim 5, it is characterised in that in described step S3 Absorbance estimation module, be made up of multiple absorbance estimation units, the implementation of each absorbance estimation unit is the most identical, It is implemented as follows:

A given frame has mist image I, and absorbance estimation unit calculates the absorbance of its correspondence:

Wherein, t is the absorbance corresponding to I, and Ω (x) is the localized mass centered by x of a size of 15 15, and β is that air dissipates Penetrate coefficient, β=1.0, I^valAnd I^satIt is the brightness under hsv color space and saturation respectively, α₀、α₁And α₂For linear coefficient, α₀ =0.1893, α₁=1.0267, α₂=-1.2966.

Video defogging method based on Distributed Calculation the most according to claim 6, it is characterised in that in described step S4 Atmosphere light illumination estimation module, is made up of multiple atmosphere light illumination estimation units, and the implementation of each estimation unit is the most identical, It is implemented as follows:

Wherein, A represents the atmosphere light illumination corresponding to picture frame I.

Video defogging method based on Distributed Calculation the most according to claim 7, it is characterised in that in described step S5 The generation module of image without mist, by multiple image generation units without mist, each unit all uses same method to restore without mist image, Concrete restored method is, according to having mist image I, absorbance t, atmosphere light illumination A, in conjunction with atmospherical scattering model, and can be by following formula Calculate without mist frame J, thus restore without mist image: