FR2811187A1 - Public place/road traffic/site surveillance having digital video sequences with predictive/non predictive images with adaptive time coding varying image step sizes following alert state. - Google Patents

Abstract

The video surveillance system has a digital image coder providing video sequences. Predictive and non predictive images are generated setting the adaptive time coding as a function of the surveillance state. A normal state or detection of an event is found. The adaptive system operates on step samples between images.

Description

VIDEO SURVEILLANCE PROCESS AND INSTALLATION

  WITH ADAPTIVE CODING FOR TRANSMISSION

  AND / OR STORAGE OF VIDEO SEQUENCES

  The invention relates to a method and a device for adaptive digital image coding for the transmission and storage of low-cost video sequences.

  debit, in the field of remote monitoring.

  Video surveillance is used in various fields of activity, in particular for the surveillance of people's property: site surveillance

  industrial, surveillance of public places, surveillance of road traffic.

  In all these fields of activity, the remote monitoring systems must carry out surveillance but also record sequences of video images,

  at least when an abnormal event occurs.

  It is a matter of being able to "replay" the filmed event, as on a video recorder, in order to be able to identify people and analyze the situation. These image sequences can be transmitted to the competent authorities

  who can use it as evidence of offenses.

  Digital systems combining a remote monitoring system

  and a recording VCR were offered.

  In most cases, the remote monitoring systems are thus coupled to digital or analog video recorders which permanently record the images acquired over periods ranging from 24 hours to 30 days. The disadvantages of these systems are the constraints of use (change of cassette every 12 or 24 hours, search for a particular sequence, archiving

said cassettes, etc.).

  To avoid the drawbacks of these systems, the invention provides an adaptive method and installation making it possible to transmit images and / or record sequences of images on a medium, in particular a

  hard drive, especially when an event is detected.

  US Patent 5,915,069 (SONY CORP.) Proposes a video surveillance method allowing the recording of images at different compression rates on a mass support, this compression rate being spatially adaptive by choosing from three quantization tables to increase or decrease the compression ratio, so there is no correlation between the

images.

  The basic idea of the invention is to implement a temporally adaptive coding capable of responding to the different situations encountered,

  by allowing adaptation over time, which is more efficient.

  The invention thus relates to a method of video surveillance from at least one camera coupled to a digital coder to obtain digital video sequences, as well as to means for recording and / or transmission of said sequences, characterized in that the digital coder generates type I, P and P 'images and implements temporal adaptive coding to adapt the binary coding according to a monitoring state, namely a normal state or an event detection state , said adaptation of the binary coding being effected by acting on the number AI of images P and P 'between two images I, and / or on the number AP of images P' between two images P. This adaptation can also be perform in

  acting on the rate of image acquisition.

  The method can be characterized in that in transmission mode, it implements a lower prediction rate than in recording mode, with the introduction of images I, non-predicted images P ′ and predicted images P, whereas in recording mode, the higher prediction rate is accompanied by the transmission of only I and P images, so as to obtain a higher image quality. This allows for example to use in transmission mode of low speed networks of type Numéris 64 kbits / s or 128 kbits / s and to record on hard disks of 8G

Bytes or more.

  In transmission mode, it may be advantageous to transmit only

  some of the images P 'generated.

  The invention also relates to a video surveillance installation comprising one or more cameras coupled to a digital coder for obtaining digital video sequences, as well as means for recording and / or transmission of said sequences, characterized in that the digital coder is temporally adaptive to adapt the binary coding as a function of a monitoring state, namely a normal state or state of detection of an event, said adaptation of the binary coding being effected by acting on the number AI of P and P images 'between two images K, and / or on the number AP of images P' between two images P. This adaptation can also be carried out by acting on the rhythm

acquisition of images.

  The installation can be characterized in that in transmission mode, the adaptive coder has a lower prediction rate than in recording mode, with the introduction of I images, non-predicted images P 'and images

  predicted P, possibly with transmission of only some of the images P '.

  whereas in recording mode, the higher prediction rate is accompanied by the transmission of only I and P images, so as to obtain a quality

higher image.

  Said adaptation of the binary coding can be carried out by acting on the desired quality of the images and on their rhythm. In recording mode, it is advantageous for the coder to configure said adaptation also as a function of a recording duration which takes account of the binary coding and of a storage capacity of the chosen recording medium. The adaptation of the binary coding is preferentially performs

by changing the bit rate.

  In transmission mode, it may be advantageous to transmit only

  some of the images P 'generated.

  Other characteristics and advantages of the invention will appear

  reading the description below, in conjunction with the drawings, in which:

  - Figures 1 and 2 illustrate an example of installation for implementing the method according to the invention, - Figures 3a to 3d illustrate various coding modes with bit rate reduction, - Figure 4 illustrates the decoding of coded sequences according to

Figures 3a to 3d.

  The invention in its preferred mode implements a coupled system (remote monitoring system + recording device) by proposing a coding method with rate reduction multi-cameras at double rate in order to allow image transmission at any time and / or recording sequences

  images, especially on hard disk, when an event is detected.

  According to the invention, the digital coder of each camera can independently manage the rate of coding of images as a function of the operating modes (transmission, recording, monitoring). These different operating modes are advantageously determined according to different detection methods, including image analysis methods, which make it possible to signal

  abnormal events such as intrusion, obstruction of the camera field, etc.

  The field of digital image coding has taken on considerable importance since the 1980s. Among the techniques used, some are dedicated to bitrate reduction coding (standards H261, JPEG, MPEG1, MPEG2, JPEG2000, etc.) and they are all based on the principle of transformation coding

of blocks.

  Thanks to this transformation, the algorithms can calculate parameters of spatial correlation, temporal correlation, subjective redundancy or statistical redundancy in order to strongly compress the images. The discrete cosine transform (subsequently TCD) in the JPEG, H261, MPEG1 and MPEG2 systems and the wavelet transform in JPEG2000 are the most used for

  determine the spatial correlation of an image.

  For motion pictures, these algorithms also use prediction techniques to further reduce the throughput of compressed pictures. For these algorithms (in the MPEG1, MPEG2, H261 standards), it is a question of being able to code the error between an image at a given instant and that which will have preceded, used as its prediction so as not to code, for example by estimation of movement, that changes that affect the animated sequence. These techniques are based on the extraction of the differences between two successive images by performing a subtraction of said images. In order to maintain acceptable image quality, the

  predictor is constantly updated (in a search for the optimal predictor).

  In this case, we speak of: - I images: coded intra without temporal predictor, they serve as input to the decoding process, - P images: predicted from the previous I or P image ("forward prediction ") with the use of motion compensation techniques, - B images: temporally interpolated from two previous I or P images, according to a so-called bidirectional prediction (" bidirectional

prediction ").

  The invention proposes to use these rate reduction coding techniques and to subject them to event analysis methods, in particular image analysis methods (detection of abnormal events in the image as a intrusion) in order to perform several functions, mainly the functions of

  transmission, recording and monitoring.

  The advantage of this process lies in the use of known but controlled techniques in order to obtain a bit rate reduction image coder whose rate may be different for transmission and for recording (double

  rhythm). The same coder can manage several cameras independently.

  Figure 1 shows an installation as a whole. For a digitized image IM obtained from a digital video camera CAM, an image analysis is carried out in a DET detection module. For this IM image, the DET module provides a signal SD for detection or non-detection of an abnormal event. As a function of this signal SD, a rhythm regulator RR provides an output signal SR containing the instructions for adapting the rhythm of a processor module

  COD image coding, both for recording and transmission.

  The current image IM and the timing information contained in the signal SR are transmitted to the rate reduction coding processor COD. At the end of this coding, the compressed CIM images are transmitted to a BUS bus to be recorded for example on a hard disk HDD or transmitted for example by an Ethernet network TR or both at the same time. They can also be displayed on SURV surveillance screens. The DET, RR and COD modules are advantageously part of a

  piP microprocessor which manages all of the installation's functions.

  Figure 2 shows the COD image coding processor

flow reduction.

  The COD coder generates 3 different types of images. These types of images are as follows: - Image I: intra coded image (see FIG. 3b), that is to say a full resolution image compressed with high quality by an encoder 7. This image is output of the coder COD and decompressed locally by a decompression module 9 to provide a decompressed image I DI to initialize the

  prediction made in module 10.

  - Image P ': difference image resulting from the subtraction of the image

  current IM and the decompressed predictor DI (see Figure 3c).

  - Image P: difference image (6) resulting from the subtraction of the current image IM and the decompressed predictor DI and then added to the predictor (see

figure 3d).

  The rate reduction coding module is conventionally a module based on a block transformation, for example a TCD which transforms each block of the IM image into a matrix of coefficients which are subsequently

  quantified by an entropy coding.

  An I / P / P 'sequence at output 8 of the coder COD therefore has the form illustrated in FIG. 3a,

O:

  - Delta P represents the number of images P 'between two images P, - Delta I represents the number of images P' and P between two images I. We therefore obtain at output 8 of the coder COD image sequences called IPP 'whose characteristics are determined by the signal SR. The COD coder is a predictive coder where the predictor is defined by the formula i = delta P Predictor = I + P i = O The predictor is re-initialized every Delta I image. For each camera, the coder has an independent predictor of

that of other cameras.

  The number of predictors is therefore equal to the number of cameras.

  Thus, the compressed video stream of each camera is an independent IPP 'sequence.

  Each type of image is coded according to the JPEG standard. Each predictor (image I) is coded in a unique and identical manner according to predefined and constant quantization tables. Each predictive image P and non-predictive image P 'is coded in a variable manner according to adaptive quantification tables so that all the images P or P' have a quasi-constant size adapted to the support of

transmission.

  Naturally, a predictor (image I) is larger than an image coded from a predictor (images P or P '). On average, this difference is evaluated at a factor of 3. Transmission of a predictor I is costly in transmission and display time. The weaker a predictor, the greater the number of predictive images P (AP is low or even zero) and the more the quality of the sequence is improved. The more frequently the predictor is reset (weak AI), the more

  quality of sequence flow is improved.

  Thus for transmission, the COD encoder lowers the rate of

  prediction to decrease bit rate.

  On the other hand, in recording mode without transmission, the encoder

  COD increases the prediction rate to maintain high image quality.

  The rhythm control module RR receives the event detection signal SD and the output signal SR defines the type of image that the coder COD will have to code, as well as its intrinsic quality (quality of coding or quantization for

transform coding).

  The RR module is in particular subject to detection of abnormal events by the DET detector. Such a DET detector is an automaton supplying the rhythm control module with the signal SD representative of a certain number of states corresponding to all the elements monitored (normal state or abnormal state for each camera). According to the different states represented by the output signal SI of this automaton, the coder adapts the coding rhythm by modifying the

  Delta I, Delta P parameters, and possibly the image prediction rate.

  The method according to the invention therefore proposes to temporarily adapt the IPP 'sequences in order to be able to influence the quality of the images as a function of abnormal events (for example, intrusion detection by image analysis coupled to the system). It is a question here of slaving the parameters AI and AP to one or more events, possibly graduated according to their level of significance, by best adapting the recording and transmission media. The table below gives an example of implementation of the proposed adaptation:

  Consider the surveillance of an entrance airlock of a bank branch.

  This entry airlock is characterized by a detection of human presence in the airlock (by

  example, an intrusion detection method by image analysis coupled to the system).

  Depending on the detection schedule for the Intrusion event, the meaning is not the same. The proposed process takes this constraint into account and adapts to different levels of significance:

Table I

  Events Schedule Meaning Recording AI AP of opening of images Intrusion Open Level of recording weak weak important: images with Monitoring of the quality movenne flow of people at each in the airlock entry / exit of person Intrusion Closed Level very recording minimum minimum important: images abnormal presence maximum quality No Closed Low level: No significant recording Intrusion in case of normality very No Open Moven level: Recording Significant average Intrusion in case of normality of images to

but weak quality monitoring.

  minimum lance is required because the agency is open. The values of the different levels AI and AP will be given as an example. These different distances are used by the method for a constant acquisition rate of four images per second:

Table 2

minimum Low Medium Important

72 240 1680 3600

  AI (in nb of images) 1 0 AI (in seconds) 18s 60s 420s 900s (7 minutes) (5 minutes) AP (in nb of images) 12 40 120 240 3s l1Os 30 s 60s AP (in seconds) (1 minute) L 'temporal adaptation of the structure of IPP sequences', formalized by the change in the distances AI and AP is an adaptation of the predictors over time with respect to changes in the content of the sequence of images. This temporal adaptation of the prediction makes it possible to regulate the quality level of the image streams over time by taking account of their activity. Thus, it is useless to keep important constraints on the quality, when the sequence of images is temporally stable. On the other hand, in the event of a very important event, the quality must be maximum, to allow the recognition of an individual appearing in the

scene filmed for example.

  2 5 Naturally, a "high quality" sequence requires a quantity

  binary information more important than a "low quality" sequence.

  The temporal adaptation of the prediction therefore makes it possible to optimize the management of the quality of the images recorded as a function of important or very important events; and therefore to optimize the storage volume of the images recorded on the hard disk. Thus, the quantity of information recorded is concentrated during important or very important events. For example, by

  the operation of the process for prioritizing events on an 8 Giga- hard drive

  bytes can store images from 8 cameras for 30 consecutive days.

  It should also be noted that the servo-control of the method can be carried out on the basis of certain output parameters of the rate reduction coder (by implementing algorithms for detection of intrusion by image analysis). In particular, the DET detector in Figure 1 can be powered by the encoder output (COD) (see dotted line, rather than directly by IM images: we get

  thus a completely autonomous and automatic process.

  The method according to the invention advantageously allows a temporal slaving of the prediction which does not disturb the transmission and monitoring functions of the system. The method proposed here makes it possible to get rid of such a disturbance advantageously by its multi-rhythm nature and allows a transmission of the images at any time, whatever the form of the

  IPP 'sequences and therefore whatever the values of AI and AP.

  Indeed, the temporal adaptation of the prediction makes it possible to choose the type of image to be transmitted and therefore to respect an output volume of the images of

constant transmission.

  The examples below illustrate this advantageous aspect of the method, using two examples of recording and transmission rate according to the cases set out in table 1. It will be noted that a double rate (of

  transmission and recording) from a single compressed sequence.

  Example 2 also shows that it is possible to obtain, still from a single IPP 'sequence, several transmission rates (for 2 different transmission networks, at 64 Kilo-bits per second and 128 Kilo-bits per second) and a recording rate. We can also generalize by proposing the method for n recording media and m transmission media: each medium can be managed independently at its own pace, from the same single stream of

IPP 'sequences.

EXAMPLES

  Example 1: case of a very important event and simultaneous transmission.

  Or the case of an acquisition rate of 4 images per second. In the event of a very important event, AI and AP are set at the minimum level: either for example:

AP 3,

AI = 20

  Acquisition rate 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 ...

  Time 1 second I second 1 second 1 second 1 second 1 second ...

  Rhythm I P 'Pl Pl P Pl Pl Pl P Pl P' Pl P Pl Pl P 'P P' Pl Pl Pl Pl Pl P ...

Registration

  1 5 Transmission rate I P P P P I 'P ...

  (64 Kb / s) In this case, in order to respect the constraints of the 64 kilo-bit per second digitized transmission network, only type I and P images are transmitted. This makes it possible, on the one hand, to comply with the speed constraints of the transmission network and, on the other hand, to transmit a stream of high images.

  quality at a constant rate of 1 frame per second.

  Example 2: case of an important event and simultaneous transmission.

  Or the case of an acquisition rate of 4 images per second. In the event of an important event, AI and AP are set at the low level: Either for example:

AP = 11,

AI = 70

  Acquisition rate 1 2 34 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

  Time 1 second 1 second 1 second I second 1 second 1 second ...

  Rhythm I P 'P' P 'P' P 'P' P 'P' P 'P' P P Pl P P P Pl Pl P P 'PP' P 'P. ..

Registration

  Transmission rate I P 'P' P P 'P ...

  (64 Kb / s) Transmission rate I p PPP 'PPPP' Pl P P ... rX (128 Kb / s) In this case, in order to respect the constraints of the transmission network O: digitize 64 Kilo-bits per second , the images I, P as well as certain images P 'are transmitted at a rate of 1 image per second. It will also be noted that the method applies to a support for

  transmission of 128 Kilo-bits per second which allows to obtain a rate of 2 images per second.

  During the decoding operation of the recorded images (replay

  hard drive data) or transmitted images, just

  build the predictor from the sequences. The process is always the same whatever the adaptation chosen and whatever the rates of transmission. Indeed, each image (type I, P or P ') is recognized by the decoder and decoded accordingly as the images arrive. The only constraint lies in the fact of positioning on a predictor and also obtaining all the predictive images (image P). If this is the case, just apply the reverse process of

  encoder as summarized in Figure 4.

  This type of sequence to be decoded allows, for the recorded data, all types of consultation (read, fast read, return, fast return). Another advantageous aspect of the process therefore appears, which is to be able to configure the temporal adaptation and transmission rates at will without modifying the decoder and without influencing the mode of viewing images: direct reading in connected mode or from files saved in consultation mode. The proposed method thus consists of an adaptation coder

  multi-rhythms and a unique decoder.

Claims (12)

  1. A method of video surveillance from at least one camera coupled to a digital coder to obtain digital video sequences, as well as to means for recording and / or transmission of said sequences, characterized in that the digital coder generates non-predicted type P, P and predicted P 'images, and in that it implements temporal adaptive coding to adapt the binary coding as a function of a monitoring state, namely a normal state or a state of event detection. said adaptation of the binary coding being effected by acting on the number AI of images P and P 'between two images I and / or on the number AP of images P' between two images P. 2. Method according to claim 1, characterized in that said adaptation of the binary coding is also carried out by acting on the rhythm acquisition of images.   3. Method according to claim 2, characterized in that in recording mode, the coder configures said adaptation also as a function of a recording duration which takes account of binary coding and of a capacity of   storage of the chosen recording medium.   4. Method according to one of the preceding claims, characterized in   what in transmission mode, it implements a lower prediction rate than in recording mode, with introduction of I, P and P 'images, while in recording mode, the higher prediction rate s' accompanies the transmission of only I and P images, so as to obtain a higher image quality than in transmission.   5. Method according to one of the preceding claims, characterized in   that in transmission mode, only some of the images P 'generated are transmitted.   6. Method according to one of the preceding claims, characterized in   that the adaptation of the binary coding is carried out by modifying the bit rate.   7. Installation of video surveillance comprising one or more cameras coupled to a digital coder to obtain digital video sequences, as well as means for recording and / or transmission of said sequences, characterized in that the digital coder generates I images, non-predicted images P and predicted images P 'and is temporally adaptive to adapt the binary coding as a function of a monitoring state, namely a normal state or detection state of an event, said temporal adaptation of the binary coding s 'effecting by acting on the number AI of images P and P' between two images I, and / or on the number AP of image P 'between two images P. 8. Installation according to claim 7, characterized in that said adaptation of the binary coding is also carried out by acting on the rhythm acquisition of images.   9. Installation according to claim 8, characterized in that in recording mode, the recording is also configured as a function of a recording duration which takes account of the binary coding chosen and the capacity of   storage of the chosen recording medium.   10. Installation according to one of claims 7 to 9, characterized in that   that in transmission mode, the adaptive coder has a lower prediction rate than in recording mode, with introduction of I, P and P 'images, while in recording mode, the higher prediction rate is accompanied transmission of I and P images only, so as to obtain higher image quality.   11. Installation according to one of claims 7 to 10, characterized in   that in transmission mode, only some of the images P 'generated are transmitted.   12. Installation according to one of claims 7 to 11, characterized in that   that the adaptation of the binary coding is carried out by modifying the bit rate.