JPH10162129A - Image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processor equipped with a means which speeds up the whole process including transfer of an intermediate result between processors as a parallel type image processor. SOLUTION: It is made possible to transfer among N sets of processors 1, 1b, and 1c by connecting the processors 1a, 1b, and 1c by unidirectional ring-shaped buses 10a, 10b, and 10c, and the image processor is provided with an image input control part 20 which performs control so that time series successive images are inputted sequentially to local memories 2a and 2a', 2b and 2b', and 2c and 2c' connected to the processors 1a, 1b, and 1c in the order along the ring. Consequently, all processes of input of an image, preprocessing, transfer of the preprocessed image, detection of a moving body, and measurement are completed by one processor in a T×N time to enable the real-time processing of the time series of images, which are inputted in real time.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for processing images at high speed, and more particularly to a monitoring system for crime prevention and disaster prevention of buildings and plants, and an obstacle detection system for autonomous vehicles. Etc.,
The present invention relates to an image processing apparatus that performs real-time processing of a moving image for detecting a moving object and measuring its shape, position, moving speed, and the like using an image.

[0002]

2. Description of the Related Art Conventionally, there are a pipeline system and a parallel system as general systems of an image processing apparatus for processing an image at high speed. The pipeline method is configured by connecting a plurality of dedicated operation units that perform image processing operations in parallel on local neighborhood areas in an image in a pipeline fashion, and sequentially input images are processed in a pipeline and processed in real time. It is a method to make it.
As a device using this method, an LSI dedicated to pipeline operation such as a relatively small spatial filter of about 3 × 3 has been put to practical use.

The parallel method is a method for reducing the processing time by distributing the processing load to a plurality of processors and performing parallel processing. Methods for distributing the load include space division and time division. In the former, the image to be processed is spatially divided into a plurality of regions, and the divided partial images are input to each processor to perform parallel processing. In the latter case, in the case of moving image processing, a time-series image is temporally divided, and an image at each time is input to each processor as a processing unit to perform parallel processing.

Here, the configuration of an image processing apparatus based on the parallel system is disclosed in, for example, Japanese Patent Publication No. 3-51024 and Japanese Patent Publication No. 6-164606.
There has been a method of connecting a plurality of processors in parallel to an input / output dedicated image bus. According to such a method, an image can be input in real time from a video camera or the like while being distributed to each processor, and the processing result partially obtained by each processor can be output as one image.

[0005]

In a monitoring system for the purpose of crime prevention and disaster prevention of buildings and plants, an obstacle detection system for autonomous vehicles, etc., detection of a moving object by using an image, its shape and position. In the image processing apparatus for measuring the moving speed and the like, it is necessary to execute image processing of a complicated processing algorithm on a moving image in real time.

However, in the above-mentioned pipeline method, the processing content is limited to a simple spatial filter, and it is not possible to cope with complicated processing algorithms such as detection and measurement of a moving object.

[0007] In the parallel system, in order to detect and measure a moving object, it is necessary not only to divide and input an image to each processor to disperse the load, but also to obtain an image obtained by another processor in an intermediate process. It is necessary to use the intermediate results obtained, or to integrate the processing results of the processors. However, the conventional parallel image processing method as described above does not provide a means for speeding up the entire processing including the transfer of the interrogation result between the processors.
It has been difficult to execute moving image processing such as measurement in real time.

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and in an image processing apparatus of a parallel system, an image processing apparatus having means for speeding up the entire processing including transfer of an intermediate result between processors. It is intended to provide.

[0009]

According to the present invention, there are provided N (N is 2)
The above integers), a local memory connected to each of the processors, and image input means for externally inputting an image to each of the local memories. Are connected by a unidirectional ring-shaped bus, and data transfer means for enabling data transfer between the processors, and in order along the ring, to the local memories connected to the processors, And image input control means for controlling to successively input successive time-series images.

(Operation) According to the above configuration, continuous time-series images are sequentially input to N processors in the order along the ring. Then, the image input to each processor is subjected to preprocessing and the like, and is sequentially transferred to a processor connected downstream of the ring. Each processor:
A moving object is detected and measured using an image preprocessed by itself and a processed image input from an upstream processor, that is, a plurality of images that are continuous in time series.

For example, assuming that the input period of a time-series image is T,
Assuming that N processors are sequentially input, the input cycle of one processor is T × N time. Therefore, all processing of image input, pre-processing, transfer of the pre-processed image, and detection / measurement processing of the moving object are completed in one processor within T × N time, so that the time series input in real time Real-time processing of images becomes possible.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. Here, a moving object measurement process is taken as an example of a moving image process, and an embodiment thereof will be described in detail with reference to the drawings. First, the configuration of the image processing apparatus of the present invention will be described with reference to FIG.

FIG. 1 is a block diagram showing a configuration of an image processing apparatus according to an embodiment of the present invention. In FIG.
1a, 1b, and 1c are processors, and N processors (N is an integer of 2 or more) are provided. 2a, 2a ', 2b, 2
b ', 2c, 2c' are local memories connected to the processors 1a, 1b, 1c. Of these, 2a, 2
b and 2c are image memories having an image input port,
2a ', 2b' and 2c 'are general-purpose memories.

Reference numeral 3 denotes an image camera, 4 denotes an image input unit, 5 denotes an image bus, and these constitute an image input means.
Reference numerals 10a, 10b, and 10c denote ring-shaped buses that provide unidirectional high-speed data transfer means between the processors 1a, 1b, and 1c.

Reference numeral 20 denotes an image input control unit, which includes, in the order along the ring, local memories 2a, 2a ', 2b, 2b', 2c, 2c, 2c connected to the processors 1a, 1b, 1c.
Provided is an image input control means for controlling a continuous time series image to be sequentially input to c ′.

Next, the flow of the moving object measurement process in the embodiment will be described with reference to FIG. FIG. 2 is a diagram showing a flow of processing when the moving object measurement processing in the embodiment is performed in parallel by three processors. The squares in the figure represent tasks, the vertical position of each task represents a processor that executes the task, and the horizontal position represents the time at which the task is executed. The arrows connecting the tasks represent the flow of data, and particularly the bold arrows represent data transferred between processors using a ring bus.

As shown in FIG. 2, the moving object measurement processing is roughly classified into four tasks T0, T1, T2, and T3. T0 is an initialization in which data initialization including a reference image is performed. T1 is an image input, and T2 is a pre-processing of the image. The image input at T1 is subjected to a spatial filter such as noise removal. T3 performs detection and measurement of the moving object by inputting the reference image and the result of T2. Here, T2
Is the next reference image.

First, an initialization task T0 is executed by the processor 1a.
Is executed, the three tasks of a series of the image input task T1, the preprocessing task T2, and the moving object detection task T3 are assigned as a set by the processors 1a, 1b, and 1c in order, and the tasks overlap. Is executed.

Next, the operation of the embodiment will be described. The image input by the image camera 3 is input to the image input unit 4 as a video signal. The image input unit 4 outputs an image clock, a vertical synchronization signal, and a timing signal of a horizontal synchronization signal based on the video signal, and sequentially outputs digitized image data to the image bus 5 in synchronization with these timing signals.

Here, as shown in the timing chart of FIG. 3, the image input control section 20 receives the vertical synchronizing signal 22 of the image from the image bus 5 and synthesizes the frame write inhibit signals 21a, 21b and 21c. Output to the local memories (image memories) 2a, 2b and 2c, respectively. Here, the frame write inhibit signals 21a, 21b, 21c
Is High, writing of image data to the corresponding local memories (image memories) 2a, 2b, 2c is prohibited.

Each of the processors 1a, 1b and 1c executes an image input task T1 and waits for completion of image input. The local memories (image memories) 2a, 2b, 2c have corresponding frame write inhibit signals 21a, 21b, 21c set to Lo.
During the period w, image data is sequentially written in synchronization with the timing signals of the two image buses 5.

When the input of the image is completed, each processor 1
a, 1b, 1c execute the pre-processing task T2, and locally connect local memories (image memories) 2a, 2a
b, 2c, perform processing such as a spatial filter, and write the results to the local memories 2a ', 2b', 2c ', and at the same time, ring buses 10a, 10b, 10c.
To transfer the results to the local memories 2b ', 2c', 2a 'of the downstream processors 1b, 1c, 1a.

Next, after the completion of the pre-processing task T2, the processors 1a, 1b, 1c are connected to the upstream processors 1c, 1c.
The mobile object measurement task T3 is executed after the completion of the input of the pre-processing result from a and 1b. The result of the preprocessing task T2 corresponding to the image of the current frame and the result of the preprocessing task T2 corresponding to the image of the immediately preceding frame are read from the local memories 2a ', 2b', and 2c '.

As described above, in the present embodiment, when the input cycle time of one frame of an image is T, as shown in FIG.
By distributing the load by time division by one processor, the input period of the image per processor can be set to T
It can be three times as long as. In other words, the present apparatus can perform real-time processing continuously for a complicated moving image processing that requires a maximum of 3T time for one processor in a cycle T time.

[0025]

As described above, according to the present invention, N
Image processing apparatus comprising at least (N is an integer of 2 or more) processors, local memories connected to each of the processors, and image input means for externally inputting an image to each of the local memories. In the above, each of the processors is connected by a unidirectional ring-shaped bus, and data transfer means for enabling data transfer between the respective processors; and each of the local processors connected to the respective processors in an order along the ring. By providing image input control means for controlling the memory so that successive time-series images are sequentially input, high-speed transfer of intermediate results between processors, which was difficult with conventional parallel image processing, is provided. In addition to this, there is an effect that real-time processing of moving image processing can be performed by distributing a load by time division.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an image processing apparatus according to an embodiment of the present invention.

FIG. 2 is a view showing a processing flow when moving object measurement processing is performed in parallel by three processors in the embodiment.

FIG. 3 is a timing chart showing an input / output relationship of each signal of an image input control unit in the embodiment.

[Explanation of symbols]

 1a, 1b, 1c Processor 2a, 2b, 2c Local memory 2a'2b ', 2c' Local memory 3 Image camera 4 Image input unit 5 Image bus 10a, 10b, 10c Ring bus 20 Image Input control units 21a, 21b, 21c: frame write inhibit signal 22: vertical synchronization signal

Claims (1)

[Claims] 1. N processors (N is an integer of 2 or more), local memories connected to the respective processors, and image input means for externally inputting an image to each of the local memories. At least an image processing apparatus, comprising: a data transfer unit that connects the processors with a unidirectional ring-shaped bus to enable data transfer between the processors; An image processing apparatus comprising: image input control means for controlling a continuous time-series image to be sequentially input to each of the connected local memories.