CN110175562A - A kind of gpu acceleration pedestrian detection algorithm based on parallel computation - Google Patents

Abstract

The invention discloses a gpu-accelerated pedestrian detection algorithm based on parallel computing, which includes a data processing module, a data calculation module and a data output module. The data collection needs to be collected before the data processing, and the data collection includes a pedestrian data collection device. The data processing module includes a central processing unit, the data output end of the central processing unit is connected with a graphics processor, the data output end of the graphics processor is bidirectionally connected with the data input end of the central processing unit, and the data calculation module is Including the cumulative people flow DA core calculation and the total people flow statistics SA, the data output end of the people flow DA core calculation is connected to the data input end of the total people flow statistics SA in one direction. The GPU-accelerated pedestrian detection algorithm based on parallel computing processes a large amount of unified data through the GPU, and uses graphics-based numerical calculations as the core to perform parallel numerical calculations on pedestrian detection data, which improves processing efficiency and facilitates rapid detection.

Description

A GPU-accelerated pedestrian detection algorithm based on parallel computing

technical field

The invention relates to the field of computer vision and image processing technology, in particular to a GPU-accelerated pedestrian detection algorithm based on parallel computing.

Background technique

After years of exploration and development, computer vision has been applied in finance, machining, transportation, entertainment, medical care, security, military and many other fields, creating irreplaceable value. In the field of computer vision, target detection is an extremely important research branch. The present invention focuses on a very important and extremely valuable topic in target detection—pedestrian detection. The research basis and premise of tracking and other topics, an excellent pedestrian detection algorithm can provide strong support and guarantee for many of the above research topics, pedestrian detection algorithm, as the name suggests, is to use a pre-trained model for a given picture or video Detect it and mark the location of the pedestrian in the picture or video. This is very valuable in real life. For example, if the pedestrian detection algorithm is used in assisted driving, the vehicle can automatically identify the pedestrians in front, obtain the precise distance from the target, and avoid pedestrians in time. In case of an alarm, it can actively participate in braking, making driving easier and making travel safer and more at ease; the monitoring system integrates algorithms such as pedestrian detection, which can detect the location of pedestrians in time and monitor their behavior. Analysis, if abnormal behaviors such as staying too long are found, the security personnel can be alerted in time to prevent problems before they happen, and so on.

Existing pedestrian detection algorithms do not use graphic values as the calculation core, but simply calculate pedestrian data, and cannot process large-scale data with a highly unified type and no dependence on each other. For graphic or non-graphic highly parallel values Computation could not be processed.

Contents of the invention

1. Technical problems to be solved

The purpose of the present invention is to solve the problem that the existing pedestrian detection algorithm does not take the numerical value of graphics as the calculation core, but simply calculates the pedestrian data, and cannot process large-scale data with a highly unified type and no dependence on each other. For graphics or It is a problem that non-graphics highly parallel numerical calculations cannot handle, and a GPU-accelerated pedestrian detection algorithm based on parallel computing is proposed.

2. Technical solution

In order to achieve the above object, the present invention adopts the following technical solutions:

A gpu-accelerated pedestrian detection algorithm based on parallel computing, including a data processing module, a data calculation module and a data output module, before the data processing needs to be collected and collected, and the data collected includes a pedestrian data collection device, collected by a camera, collected The data is detected by the picture detector, and the collected pedestrian picture data is screened out. The data processing module includes a central processing unit, and the data processing module initializes the collected pedestrian picture, and outputs the initialized pedestrian picture data. The data output end of the central processing unit is connected with a graphics processor, and the data output end of the graphics processing unit is bidirectionally connected with the data input end of the central processing unit. People flow statistics SA, the data output end calculated by the people flow DA core is connected unidirectionally with the data input end of the total people flow statistics SA, and the data output module includes a receiver and a transmission device, and the data output end of the receiver One-way connection with the data input terminal of the transmission device.

Preferably, the pedestrian data collection device includes a picture collector and a picture detector, the picture collector can collect through a camera, the collected data is detected by the picture detector, and the collected pedestrian picture data is screened out.

Preferably, the data processing module initializes the collected pedestrian pictures, sets the pedestrian flow threshold DT for the initialized pedestrian picture data, and sets the detection threshold NT through detection, and sets the total threshold through the data processing module at the same time. The flow of people SA is the maximum value.

Preferably, the data calculation module receives the data processed by the data processing module, calculates the cumulative pedestrian flow DA kernel, and makes statistics on the data output to obtain the total pedestrian flow.

Preferably, the data output module outputs the pedestrian flow data through the data output device when the accumulated pedestrian flow DA value is less than the set pedestrian flow threshold DT.

Preferably, the detected data is compared with the set people flow threshold DT by comparing the total people flow SA, and when the total people flow SA is less than the people flow threshold DT, the data output is set to the data output module.

Preferably, when the pedestrian flow data is output, the detected positions are marked to facilitate calculation and processing.

Preferably, while setting the total flow of people, it is necessary to set a minimum storage and processing unit, and process the detected data in parallel.

Preferably, said transport equipment includes line matrix printers and numerically controlled plotters and other scanning devices.

3. Beneficial effect

Compared with the prior art, the present invention has the advantages of:

(1) In the present invention, through camera collection, the data of collection is detected by picture detector, screens out the pedestrian picture data that gathers, and data processing module carries out initialization process to the pedestrian picture that gathers, and the pedestrian picture data that initialization completes Output, set the threshold of people flow DT, and set the detection threshold NT through detection, and set the total flow of people SA as the maximum value through the data processing module. When setting the total flow of people, it is necessary to set the minimum storage and processing unit. The data is processed in parallel, the data calculation module receives the data processed by the data processing module, calculates through the DA kernel of the cumulative pedestrian flow, and counts the data output to obtain the total pedestrian flow, and the data output module will reduce the cumulative DA value of the pedestrian flow to less than the set value When the pedestrian flow threshold DT is set, the pedestrian flow data is output through the data output device, and the detected data is compared with the set pedestrian flow threshold DT by comparing the total pedestrian flow SA. When the total pedestrian flow SA is less than the pedestrian flow threshold DT, the data The output is located in the data output module.

(2) Calculate the thread number to receive data, set the C value as the boundary, and compare the C value with the number of pixels. When the C value is greater than the number of pixels, calculate the settlement. If the C value is less than the number of pixels, proceed to the next step to calculate zw(c) value, the calculated value is calculated by calculating the maximum slope pixel zw(n) value, and by comparing zw(c) with zw(n), when zw(c) is greater than zw(n), it is directly output and the calculation ends , when zw(c) is less than the value of zw(n), calculate the transfer flow of people MV, increase the cumulative flow of people and zw(n), output data, end the calculation, when the value of C is not the boundary, increase the cumulative flow of people, Resets the C value to zero and outputs the data to end the calculation.

Description of drawings

Fig. 1 is the structural representation of a kind of gpu accelerated pedestrian detection algorithm based on parallel computing that the present invention proposes;

FIG. 2 is a schematic diagram of the algorithm structure of a GPU-accelerated pedestrian detection algorithm based on parallel computing proposed by the present invention.

FIG. 3 is a block diagram of a GPU-accelerated pedestrian detection algorithm based on parallel computing proposed by the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention.

In describing the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", " The orientation or positional relationship indicated by "outside", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, so as to Specific orientation configurations and operations, therefore, are not to be construed as limitations on the invention.

Example 1:

Referring to Figure 1-3, a gpu-accelerated pedestrian detection algorithm based on parallel computing includes a data processing module, a data calculation module and a data output module. Data collection needs to be performed before data processing. The data collection includes a pedestrian data collection device and a data processing module. Including a central processing unit, the data output end of the central processing unit is connected to a graphics processor, and the data output end of the graphics processor is bidirectionally connected with the data input end of the central processing unit. Traffic statistics SA, the data output terminal calculated by the people flow DA core is connected to the data input terminal of the total people flow statistics SA in one direction. The data output module includes the receiver and the transmission device, the data output terminal of the receiver and the data input terminal of the transmission device One-way connection.

In the present invention, the pedestrian data acquisition device includes a picture collector and a picture detector. The picture collector can be collected by a camera, and the collected data is detected by the picture detector, and the collected pedestrian picture data is screened out. Collector and picture detector, the picture collector can be collected by the camera, the collected data is detected by the picture detector, and the collected pedestrian picture data is screened out, the data calculation module receives the data processed by the data processing module, and accumulates the pedestrian The flow DA kernel calculates and counts the data output to obtain the total pedestrian flow. The data output module outputs the pedestrian flow data through the data output device when the accumulated pedestrian flow DA value is less than the set pedestrian flow threshold DT. The detected data passes through Compare the total people flow SA with the set people flow threshold DT, when the total people flow SA is less than the people flow threshold DT, the data output is set to the data output module, and the detected data is compared with the total people flow SA and the set people flow threshold DT for comparison, when the total pedestrian flow SA is less than the human flow threshold DT, the data output is set to the data output module, and the detected position will be marked when the pedestrian flow data is output, which is convenient for calculation and processing. When setting the total pedestrian flow, it is necessary to set The smallest storage and processing unit, parallel processing of detected data, transmission equipment includes line printers, numerical control plotters and other scanning devices.

In the present invention, through camera collection, the collected data is detected by a picture detector, and the collected pedestrian picture data is screened out. The data processing module initializes the collected pedestrian picture, outputs the initialized pedestrian picture data, and sets People flow threshold DT, and set the detection threshold NT through detection, and set the total people flow SA as the maximum value through the data processing module. When setting the total people flow, it is necessary to set the minimum storage and processing unit, and carry out the detection data Parallel processing, the data calculation module receives the data processed by the data processing module, calculates through the DA core of the accumulated people flow, and counts the data output to obtain the total pedestrian flow, and the data output module will make the DA value of the accumulated people flow less than the set people flow When the threshold is DT, the pedestrian flow data is output through the data output device, and the detected data is compared with the set people flow threshold DT by comparing the total people flow SA. When the total people flow SA is less than the people flow threshold DT, the data output is set to data The output module calculates the thread number to receive data, sets the C value as the boundary, and compares the C value with the number of pixels. When the C value is greater than the number of pixels, the calculation is settled, and the C value is smaller than the number of pixels for the next step of calculating zw(c) value, the calculated value is calculated by calculating the maximum slope pixel zw(n) value, and by comparing zw(c) with zw(n), when zw(c) is greater than zw(n), it is directly output and the calculation ends , when zw(c) is less than the value of zw(n), calculate the transfer flow of people MV, increase the cumulative flow of people and zw(n), output data, end the calculation, when the value of C is not the boundary, increase the cumulative flow of people, Resets the C value to zero and outputs the data to end the calculation. The GPU-accelerated pedestrian detection algorithm based on parallel computing processes a large amount of unified data through the GPU, and uses graphics-based numerical calculations as the core to perform parallel numerical calculations on pedestrian detection data, which improves processing efficiency and facilitates rapid detection.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, any person familiar with the technical field within the technical scope disclosed in the present invention, according to the technical solution of the present invention Any equivalent replacement or change of the inventive concepts thereof shall fall within the protection scope of the present invention.

Claims (9)

1. a kind of gpu based on parallel computing accelerates the pedestrian detection algorithm, comprises data processing module, data calculation module and data output module, is characterized in that, needs to collect before described data processing, and described data acquisition comprises pedestrian data acquisition device , the data processing module includes a central processing unit, the data output end of the central processing unit is connected with a graphics processor, the data output end of the graphics processor is bidirectionally connected with the data input end of the central processing unit, and the data calculation Comprise cumulative people flow DA kernel calculation and total people flow statistics SA in the module, the data output terminal of described people flow DA kernel calculation is connected with the data input end of total people flow statistics SA one-way, and described data output module includes receiver and The transmission device, the data output end of the receiver is unidirectionally connected with the data input end of the transmission device. 2. a kind of gpu acceleration pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, described pedestrian data collection device comprises picture collector and picture detector, and described picture collector can be collected by video camera, The collected data is detected by a picture detector, and the collected pedestrian picture data is screened out. 3. a kind of gpu-accelerated pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, described data processing module carries out initialization processing to the pedestrian picture that gathers, and the pedestrian picture data data that initialization completes, sets Threshold DT of the flow of people, and set the detection threshold NT through detection, and set the total flow of people SA as the maximum value through the data processing module. 4. a kind of gpu acceleration pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, described data calculation module receives the data after data processing module processing, calculates by accumulative flow of people DA kernel, data The output is counted to obtain the total flow of pedestrians. 5. a kind of gpu acceleration pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, when described data output module is less than the people flow threshold value DT of setting when cumulative flow of people DA numerical value, by data output device will Pedestrian flow data output. 6. a kind of gpu-accelerated pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, the data after described detection is compared with the people flow threshold value DT of setting by comparing total people flow SA, when total people When the flow rate SA is less than the people flow threshold DT, the data output is set to the data output module. 7. a kind of gpu accelerated pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that. 8. a kind of gpu-accelerated pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, said minimum storage and processing unit needs to be set when total flow of people is set, and the data detected is processed in parallel . 9. a kind of gpu accelerated pedestrian detection algorithm based on parallel computing according to claim 1, is characterized in that, described transmission equipment comprises line matrix printer and numerical control plotter and other scanning devices.