JP5982557B2 - Video surveillance system and image search system - Google Patents

Description

The present invention relates to a video surveillance system, and more particularly to a technique for performing detection processing such as face / person detection on video acquired by a camera.

  In conventional large-scale surveillance systems, specific objects such as faces, people, and vehicles are detected from the images of surveillance cameras installed in various locations, and feature quantities are extracted from the detected objects and registered in the database.・ We were managing. As a result, it was possible to search for a specific person using the accumulated object features and to instantly analyze the movement trajectory of the specific person.

  However, the conventional system has a problem in that the recognition accuracy deteriorates depending on the installation situation because the illumination environment, the depression angle, and the like are greatly different for each of the plurality of monitoring cameras on the network connected to the system. For example, the face detection accuracy may be deteriorated for an image of a camera installed at a large depression angle or a camera installed at a position where the surrounding illumination fluctuation is severe. In addition, when the face orientation of a person reflected between the cameras is different, the face orientation is different even if a face search is performed from the camera 2 having a different depression angle based on the feature amount of the face photographed by a certain camera 1. It may be difficult to find the face of the same person.

  On the other hand, Patent Document 1 proposes a method of stabilizing illumination variation by selecting an image processing method based on a table prepared in advance according to the temperature of the use environment. Similarly, Patent Document 2 aims at stabilization by photographing a face using two cameras and switching a face authentication threshold according to the face situation obtained from each image.

JP 2012-134875 A JP 2008-108243 A

P. Viola and M.M. Jones, “Rapid Object Detection using Boosted Cascade of Simple Features”, CVPR2001

  However, in Patent Document 1, there is a problem that the system is complicated because it must be measured using another means such as temperature. In Patent Document 2, two cameras must be installed at the same position, which increases the cost. On the other hand, it may be possible to manually set different recognition processes for each camera, but this requires specialized knowledge. If the number of cameras is more than a few hundred or so, this setting takes a lot of time. It was necessary.

  In order to solve the above problems, for example, the configuration described in the claims is adopted. The present application includes a plurality of means for solving the above-described problems. To give an example, a monitoring system includes a camera and a detection unit that detects an object from an input image captured by the camera. A feature amount extraction unit that extracts a feature amount of the target object, a storage unit that accumulates the input image, the target object, and the feature amount, a selection control unit that controls the detection unit and the feature amount extraction unit, a detection unit, An evaluation unit that evaluates a result from the feature amount extraction unit, and the selection control unit selects a detection method in the detection unit and an extraction method in the feature amount extraction unit based on the output from the evaluation unit It is characterized by that.

  According to the present invention, an optimum detection / extraction process can be automatically selected for each camera under different circumstances for each of a plurality of surveillance cameras.

It is a figure which shows the outline | summary of Example 1 of this invention. It is a figure which shows the system configuration | structure of this invention. It is a flowchart which shows the flow of the detection process in an object detection part. It is a figure which shows the outline | summary of the detection system in an object detection part. It is a flowchart which shows the flow of the feature extraction process in a feature extraction part. It is a figure which shows an example of the recognition result output in a recognition result output part. It is a figure which shows the structure of an evaluation part. It is a flowchart which shows the flow of collection of the evaluation image which an image collection part performs. It is a figure which shows the outline | summary of the recognition result determination in a result determination part. It is a figure which shows an example of the statistical analysis result in a statistical analysis part. It is a figure which shows an example of the system selection result in a selection control part. It is a figure which shows an example of the management screen of the recognition part in a display part. It is a figure which shows an example of the screen which displays the result of having performed the image search in a display part. It is a figure which shows an example of the result performed regarding the position and size on the image of the object which detected the statistical analysis in a statistical analysis part. It shows an example of a special condition time system control table in Embodiment 3 of the present invention.

  FIG. 1 shows an outline of the present invention and is a diagram showing an example of a monitoring system using the present invention. This system has one or a plurality of cameras 110 a to 110 b, and these cameras are connected through a communication infrastructure 120. The communication infrastructure 120 is a LAN or a video transmission cable, and the video of each camera is transmitted through the communication infrastructure 120. The image recognition unit 130 is a processing unit placed on a server or a CPU in the camera, and performs a recognition process such as face detection on an image obtained from the camera 110, and the recognition result is stored in a feature amount database through the communication infrastructure 120. 140.

  The feature quantity database 140 has a function of storing and managing feature quantities obtained from the image recognition unit 130. Further, it has a function of searching for similar feature value data from a plurality of accumulated feature values. The image feature amount stored in the feature amount database 140 is associated with an image stored in the image database 150 using an ID or the like. The image database 150 stores image data obtained from each camera 110.

  The image data stored here is added with time information, camera information, and the like, and has a function of instantaneously searching for a predetermined image based on such information. The management unit 160 has a function of managing the image recognition unit 130 such as control of the control method set by the selection control unit. The display unit 170 searches and browses image data and the like stored in the image database, and manages the state of the recognition unit 130 through the management unit 160. The sensor 180 is a variety of sensors such as an infrared sensor and an RFID reader, and sensor information can be acquired by the image recognition unit 130 through the communication infrastructure 120. However, the sensor 180 is not an essential configuration.

  In the image recognition unit 130, the image recognition method performed by the object detection unit 132 and the feature extraction unit 133 is first determined by the selection control unit 131 for the image obtained from the camera 110. The object detection unit 132 has a plurality of detection methods 1 to N, and performs object detection according to the method selected by the selection control unit 131. Similarly, the feature extraction unit 133 also extracts the feature amount of the object from the object obtained at 132 using the extraction method determined by the selection control unit 131.

  The recognition result output unit 134 sends the obtained object feature information to the feature database 140 via the communication infrastructure 120. Moreover, it sends to the evaluation part 135 according to a condition. The evaluation unit 135 evaluates various methods of the object detection unit 132 and the feature extraction unit 133 and sends the evaluation results to the statistical analysis unit 136. The statistical analysis unit 136 statistically analyzes a plurality of obtained evaluation results to determine the optimum object detection method and feature extraction method for each camera or for each camera time zone, and to select a method. Create and update tables regularly. The selection control unit 131 performs method selection according to the method selection table of the statistical analysis unit 136. With the above configuration, object detection and feature extraction processing can be performed using an optimal recognition method determined in advance for each camera, and optimal recognition processing according to installation conditions can be automatically performed.

  FIG. 2 is a diagram showing a system configuration of the present invention. The image recognition unit 130 is software that operates on the server computer 210. The server computer 210 includes an I / F 211 that transmits / receives information to / from a communication base, a CPU 212 that performs processing of the image recognition unit 130, a memory 213, and an HDD 214 that stores information. Similarly, the management unit 160 is executed on the CPU 212, and the management information is stored in the HDD 214. The display unit 170 is realized in the client computer. The present invention is implemented on the above system configuration. One server computer or a plurality of server computers may be provided. The client computer 170 may be realized on the same device.

  FIG. 3 is a diagram illustrating an example of a processing flow of the object detection unit 132. Here, an example in which face detection is performed as an object is shown, but human or vehicle detection may be performed according to a target to be detected. Preprocessing is performed on the image obtained in S31 in S32. For example, filter processing such as image scale change, super-resolution, smoothing, and edge enhancement are candidates. In step S33, a moving object detection process is performed to extract a moving object region. In S34, face detection is performed on the moving object region obtained in S33 to detect a face. Face detection can detect a face from an image by using a discriminator constructed by machine learning in advance by using a method described in Non-Patent Document 1, for example.

  An example of this discriminator is shown in FIG. A discriminator is constructed in advance for each face direction, such as a front face discriminator capable of detecting a front face, a side face discriminator capable of detecting a left side face, and a downward face discriminator capable of detecting a downward face. A classifier may be prepared for each rotation of the face, or a classifier may be prepared for each object such as a person or a vehicle. At the time of execution, detection processing is performed using a discriminator selected by the selection control unit 131 from among the plurality of discriminators. In S35, an erroneous detection determination is performed on the detection result obtained in S34, and the detection result is output in S36. Execution / non-execution of processing in each step and selection of a plurality of methods are performed based on the result determined by the selection control unit 131.

  FIG. 5 is a diagram illustrating an example of the flow of object feature amount extraction processing in the feature extraction unit. In S51, the object detection result obtained by the object detection unit 132 is received. In S52, the object o is selected from all the detected objects, and the following processes of S53 and S54 are performed. In S53, pre-processing is performed on the detected face image. Pre-processing includes normalization of illumination, image enlargement by super-resolution, and image position correction by detecting the position of a facial organ such as the eyes and nose and mouth.

  Next, feature amount extraction is performed on the face image on which the preprocessing has been performed in S54. In the feature amount extraction, multi-dimensional D image feature amounts are extracted by PCA (principal component analysis), Gabor Filter, HOG feature amount calculation, or the like. As this feature amount extraction, a method for extracting D1-dimensional feature values from the entire head, a method for extracting D2-dimensional feature values only from the face region excluding the region with hair, and the pre-processing of S54 are used. For example, a technique for extracting a local feature amount from the periphery of the given facial organ position in a three-dimensional manner is prepared, and this is switched in the designation of the selection control unit 131. Here, D1, D2, and D3 <D.

  In S55, it is determined whether or not feature amounts have been extracted for all detected objects. If there is an unextracted object, the process returns to SE2, and if not, the process moves to S56. In S56, the obtained feature quantities of all objects are output. Accordingly, for example, when the face tends to appear small on the camera, the D1 dimensional feature value of the entire head is selected, and when the face appears large, selection is performed such that the overall D1 dimension + face part D2 dimensional feature value is extracted. By extracting a plurality of feature quantities in this way, a search is performed using only the D1-dimensional whole head feature quantity between a camera with a small face and a camera with a large face. It is possible to perform a search using the feature amount.

  FIG. 6 is a diagram illustrating an example of an output result of the recognition result output unit 134. The recognition result output unit 134 adds the camera number and time information to the object information obtained by the object detection unit 132 and the feature extraction unit 133 to obtain a recognition result and registers it in the feature amount database 140. The information required for the recognition result includes the photographed camera number, the frame ID unique to the photographing time, the detection processing method and feature extraction method selected by the selection control unit 131, the detection region indicating the object position on the image, D-dimensional feature value obtained from the object.

  The obtained recognition result information is accumulated as a recognition result table 610, which may be sent for each image frame, or may be sent at regular intervals. By adding information on the detection method and the feature extraction method, it is possible to manage and record the files separately for these pieces of information in the feature amount database. Thereby, the speed of data search can be increased. In addition, it is possible to search within the feature quantities obtained as a result of the same method, or to perform a search by performing a predetermined conversion between different methods, thereby improving the search accuracy.

  FIG. 7 is a diagram showing a detailed configuration of the evaluation unit 135. The image collection unit 701 acquires an evaluation image from the transmission unit 120 and stores it in the evaluation data storage unit 703. However, accumulation in the evaluation data accumulation unit 703 is performed only when specific sensor information and recognition results are acquired. The sensor information acquisition unit 702 acquires sensor information from the transmission unit, and notifies the image collection unit when specific sensor information is acquired. This sensor information is, for example, a human sensor, entry / exit card information using RFID, radio wave information using a mobile phone, or the like. If card information at the time of entry / exit is used, it is possible to roughly estimate how many people have entered the camera shooting area from where. The evaluation data storage unit 703 stores the images collected by the image collection unit 701. At this time, sensor information and recognition result information are added and recorded.

  The evaluation execution control unit 704 sends a recognition processing execution command for the image of the evaluation data storage unit 703 to the selection control unit 131. The selection control unit 131 issues a control command to execute all the object detection methods 1 to N and feature extraction methods 1 to M at the time of evaluation execution. For example, this evaluation execution control unit outputs an execution command in a time zone such as nighttime or a time zone in which no one is shown and the processing load of the CPU 212 is light.

  The recognition result for the evaluation image instructed by the evaluation execution control unit 704 is input from the recognition result output unit 134 to the recognition result acquisition unit 705 and sent to the result determination unit 705. The recognition result acquisition unit 705 also acquires a recognition result at a normal time other than at the time of evaluation execution, and notifies the image collection unit of the result when a specific recognition result is acquired. This specific recognition result is, for example, a case where the number of detected objects is large, or a case where almost no objects can be detected although there are many moving objects. Each of these is suitable as an evaluation image as an image in which many people would have been detected or an image in which a detection failure would have occurred. Therefore, the image collection unit 701 determines whether to collect this image and stores it. . The result determination unit 706 performs accuracy evaluation on the recognition results of a plurality of methods, and collectively sends the results to the statistical analysis unit 136.

  By adopting the above configuration, images suitable for evaluation can be collected periodically, and further, correct information (e.g., the position of the person and who is the person) based on the sensor information and recognition results. Etc.) can be collected automatically. Evaluation of a plurality of recognition methods can be performed on the evaluation image thus obtained without interfering with necessary recognition processing.

  FIG. 8 is a flowchart for determining image collection in the image collection unit 701. When the discrimination process is started in S81, first, in S83, it is determined whether the sensor value obtained from the sensor information acquisition unit 702 is equal to or greater than the threshold value Ss. For example, if the sensor is a human sensor, the threshold is the amount of the sensor, and if the sensor is passing information for a gate or a door using RFID or the like, the number of people passing within a certain time (for example, 3 or more) is set. . If it is equal to or greater than the threshold, the process proceeds to S86, where the current image is collected and cached in a temporary memory. If it is less than the threshold, the process proceeds to S84. In S84, the recognition value is compared with the threshold value Sr, and if it is equal to or greater than the threshold value, the process proceeds to S86, and if it is less than the threshold value, the process proceeds to S85. This recognition value is, for example, the value of the number of moving object regions or the distance between the extracted facial feature quantity and the person registered in advance. In S85, it is determined whether a predetermined time T2 seconds or more has elapsed from the time when the evaluation image was previously recorded. If it has not elapsed, the process returns to the determination process in S83, and if it has elapsed, the process proceeds to S87 to be temporarily stored. The recorded image is recorded in the evaluation data storage unit.

  After recording, the process proceeds to S82, and after waiting for T1 seconds in S82, it is continuously determined whether image collection is performed. The reason for waiting for T1 seconds is to prevent a similar image from being stored. Through the above processing, evaluation images can be accumulated one by one at a constant interval T2 seconds. Further, it is possible to collect and collect images more suitable for evaluation using sensor values and recognition values.

  FIG. 9 is a diagram illustrating an example of the determination result of the result determination unit 706. For example, in places that pass through entrance / exit gates when going to work, information about the number of people photographed by the camera and who the person is is added to the image based on the ID at the time of passage as sensor information. And recorded in the evaluation data storage unit 703. Based on this information, the detection rate of how many of the three people in the image can be detected and the distance (similarity) between the feature quantity obtained from a specific person and the registered feature quantity of that person ) As an evaluation result. As described above, the accuracy of the object detection and the feature amount can be calculated based on the sensor information.

  FIG. 10 is a diagram illustrating an example of an analysis result held by the statistical analysis unit 136. The statistical analysis unit 136 accumulates the evaluation results obtained by the evaluation unit 135 in the DB, and performs statistical analysis on the accumulated data at regular time intervals. As a result, the statistical analysis unit obtains an analysis result table 1010. Tables 1010a and 1010b show examples of analysis result tables of the cameras 110a and 110b, respectively.

  The result table includes items such as a specific time zone, weather, detection method, detection area, and feature extraction method, and the recognition result is divided into time zones, so that the selection control unit 131 recognizes each specific time zone. It is possible to switch the method. Similarly, it is possible to change the recognition method for each weather by having the weather information at that time. In the item of the detection method, the positive detection accuracy is obtained as a recognition rate for each detection method, and in the feature extraction method, the authentication accuracy for each feature amount extraction method is obtained by calculating an average from all the evaluation data. Thereby, it is possible to statistically grasp which recognition method has the highest performance in a certain time zone.

  The detection area stores an area where all detected objects exist. As a result, it is possible to grasp the region where the object will exist for each time slot. The total number of data used for statistical analysis may also be stored in this table. Thereby, the reliability of each recognition rate can be determined, and when the statistical data is small, it is possible not to select a method. In addition, a statistical analysis table may be created and output for each lighting situation, each face size, and each CPU load situation.

  FIG. 11 is an example of a method selection table of the selection control unit 131. For each camera and time zone, an object detection method and a feature extraction method are defined so that recognition processing with the highest accuracy can be performed based on the result of the statistical analysis unit 136. During execution, the processing contents of the object detection unit 132 and the feature extraction unit 133 are designated according to this table. This method control table 1110 is updated at a specific period such as every day or every week. Also, if there is a change in the camera installation position, it is reset.

  By adopting the above configuration, it is possible to perform image recognition processing on an enormous number of cameras installed at a plurality of locations constituting the monitoring system using an optimum recognition method according to each environment.

  FIG. 12 is an example in which a monitoring system management screen is displayed on the display unit 170. Management information of the monitoring system is managed by the management unit 160. For example, the system control table 1110 of each camera held by the selection control unit 131 and the analysis result table 1010 held by the statistical analysis unit are collectively managed. The display unit 170 performs processing for extracting necessary information from the management unit 160, shaping it, and displaying it in an easy-to-understand manner. These can be constructed using HTML, Flash, or the like. FIG. 12A shows an example of a processing method confirmation screen 1210. On this screen, the object detection method and the feature extraction method adopted by each camera can be confirmed. Here, it is also possible to change the method selection table of the selection control unit 131 through the management unit 160 by changing the setting of each item and pressing the “update” button. As a result, many recognition methods can be automatically determined based on the statistical analysis result, and only fine adjustment can be performed manually. In addition, by pressing the “fix” button, it becomes possible to continue using the determined method without updating in the future. By displaying the recognition result for the current image of the camera in the current image window at the lower right of the screen, the recognition method can be finely adjusted while checking the image.

  FIG. 12B is an example of the evaluation result confirmation screen 1220. Here, the evaluation result obtained by the statistical analysis unit 136 can be confirmed through the management unit 160. In addition, by selecting a method to be confirmed, it is possible to superimpose and display an evaluation image, its recognition result, correct answer information, and the like in a window at the bottom of the screen and confirm the evaluation result. The buttons at the bottom of the screen can be used to control playback such as sending and returning images, and the buttons on the right side of the screen can be used to delete evaluation images and add / correct correct information. Thereby, the quality of the evaluation image can be improved and more accurate accuracy evaluation can be realized.

  FIG. 13 is an example of a screen for performing an image search on the display unit 170. The search result display screen 1310 searches for an image showing a similar face by selecting a specific face image in the query image window and pressing a search button on the side of the window. The search command is transmitted to the feature quantity database 140, searches the feature quantity database 140 for data having a feature quantity similar to the face of the query, and outputs the data in the database in a similar order (in order of distance). . The display unit 170 draws out an image from the image database 150 based on the frame ID information in the acquired data and displays it on the screen. At this time, it is also possible to display the search result by limiting the detection method and the feature extraction method only to an image that has been processed in the same manner as the query image. It is also possible to display separately in the form of a result between “near methods” in which only the 1 and 2 methods are different, and a search result between “different methods” in which the methods are different beyond that. For example, while the same face distance is 0.1 between the same methods, the same person's face distance may be only 0.3 at least between the different methods. If this is displayed as the same search result, a face photographed by another method will be behind the search result, making it difficult to find. By displaying separately for the same method and different methods, it becomes easier to find the image search result of the same face.

  Based on the above, the monitoring system described in the present embodiment includes a camera, a detection unit that detects an object from an input image captured by the camera, and a feature amount extraction unit that extracts a feature amount of the detected object. A storage unit that accumulates an input image, an object, and a feature amount; a selection control unit that controls the detection unit and the feature amount extraction unit; and an evaluation unit that evaluates a result from the detection unit and the feature amount extraction unit; The selection control unit selects a detection method in the detection unit and an extraction method in the feature amount extraction unit based on the output from the evaluation unit.

  With these features, images suitable for evaluation can be collected periodically, and correct information (such as the person's position and who the person is) in the evaluation image is automatically based on sensor information and recognition results. Can be collected.

<Limitation of detection area>
A detection area limiting method according to the second embodiment of the present invention will be described with reference to FIG. FIG. 14A is an image of a passage taken by a certain camera 110a, and FIG. 14B is an image showing an example in which the recognition result accumulated in the statistical analysis unit 136 is drawn on the image. As shown in FIG. 14B, an object such as a face appears in a specific size for each area with respect to a limited area on the passage. Therefore, by performing statistical analysis, the object size existing for each divided area on the image can be limited as shown in FIG. FIG. 14C shows that no object appears in the area written as 0. Other than that, the minimum and maximum size of the appearing object is shown. By referring to this table by the selection control unit 131, it is possible to limit the area where the object is detected and the size of the object, and it is possible to reduce processing time and eliminate false detection.
Furthermore, it is possible to estimate the MAP at the position where the camera is installed based on the appearance distribution obtained by the statistical processing. By comparing this with map data of a facility or the like where the camera is installed, it is possible to estimate where on the map the image is taken, and to automatically estimate the current camera position.

<Evacuation guidance>
A monitoring system for special situations according to the third embodiment of the present invention will be described with reference to FIG. FIG. 15 shows an example of the special condition time system control table 1510 held by the selection control unit 131. The special condition priority method table 1510 indicates the priority of processing of the object detection unit 132 and the feature extraction unit 133 when a special condition occurs. The special condition priority system table 1510 is set manually because a large amount of data cannot be collected statistically for emergency response. The special condition priority method table 1510 has a special situation item, and the priority of the switching recognition method is assigned to each special situation such as a fire or an earthquake. The occurrence of a special situation such as a fire or an earthquake can be obtained from the sensor through the communication infrastructure 120. A fire can be detected by attaching a smoke detection sensor as the sensor 180, and an earthquake can be detected by obtaining notification of an emergency earthquake warning or sensor information of a seismic intensity meter. When the sensor information is received by the selection control unit 131, the recognition process is performed according to the special condition time method control table 1520 generated from the special condition time priority method table 1510 and the analysis result table 1010. The special condition method control table 1520 has a value obtained by multiplying the recognition rate R of each method included in the analysis result table 1010 and the priority P of the special condition priority method table 1510 above a certain threshold value. Select the method that is the largest of them. This makes it possible, for example, to focus on moving object detection without performing face detection in the event of an earthquake, or to focus on recognition processing on a camera that is important in another disaster without performing any recognition processing on one camera. It becomes. The minimum detection size included in the special condition priority method table 1510 defines the minimum size at which the object in the special situation appears in the image. By comparing this value with the minimum / maximum size of the object included in the statistical analysis unit 136, it is possible to determine how much the zoom or pan / tilt of the camera should be in a special situation. By describing this in the special condition time system control table 1520 as a camera switching value, it is possible to set an optimal camera parameter when a special situation occurs based on the statistical analysis result of each camera. For example, the entire camera can be monitored by switching a camera that has been set to zoom so that a person's face can be clearly seen in normal times so that it can be photographed at a wider angle when an earthquake occurs. In addition, if a table is prepared for each setting value of the PTZ camera as a special situation, it is possible to switch the detection method for each PTZ direction of the camera.
With the above configuration, even in special situations such as disasters such as earthquakes and congestion such as traffic jams, the optimal recognition method is implemented for each camera environment based on priorities manually set in advance and past statistics. It becomes possible.

110a, 110b ... camera,
120 ... communication infrastructure,
130: Image recognition unit,
131 ... selection control unit,
132 ... object detection unit,
133 ... feature extraction unit,
134 ... recognition result output unit,
135 ... Evaluation department,
136 ... statistical analysis section,
140 ... feature quantity database,
150 ... Image database,
160 ... management part,
170 ... display section,
180a ... sensor 210 ... server computer,
211 ... I / F,
212 ... CPU,
213 ... Memory,
214 ... HDD
610 ... Recognition result table 701 ... Image collection unit,
702 ... Sensor information acquisition unit,
703 ... evaluation data storage unit,
704 ... Evaluation execution control unit,
705 ... Recognition result acquisition unit,
706 ... result determination unit 1010 ... analysis result table 1110 ... method control table 1210 ... processing method confirmation screen,
1220 ... Evaluation result confirmation screen 1310 ... Search result confirmation screen 1510 ... Special condition priority method table,
1520 ... Special condition time system control table.

Claims (9)

A camera,
A detection unit for detecting an object from an input image photographed by the camera;
A feature amount extraction unit that extracts a feature amount of the detected object;
A storage unit for storing the input image, the object, and the feature amount;
A selection control unit that controls the detection unit and the feature amount extraction unit;
An evaluation unit that evaluates results from the detection unit and the feature amount extraction unit;
A monitoring system, wherein a selection control unit selects a detection method in the detection unit and an extraction method in the feature amount extraction unit based on an output from the evaluation unit. The monitoring system according to claim 1,
A monitoring system further comprising a statistical analysis unit that accumulates the results output from the evaluation unit, performs statistical analysis, and outputs the analysis results to the control selection unit. The monitoring system according to claim 2,
A monitoring system characterized in that a face is detected as the object. The monitoring system according to claim 2,
The storage unit accumulates the input image every specific time,
The detection unit and the feature amount extraction unit perform a plurality of detection methods and extraction methods on the input image when the processing load of the detection unit and the feature amount extraction unit is small,
The said evaluation part evaluates the result of having performed each said detection method and each said extraction method, respectively, and accumulate | stores in the said memory | storage part. The monitoring system according to claim 4,
A sensor information acquisition unit;
The monitoring system, wherein the image acquired by the sensor information acquisition unit is used by the evaluation unit. The monitoring system according to claim 2,
The monitoring system, wherein the statistical analysis unit calculates the performance of the camera for each time period from the result output from the evaluation unit. The monitoring system according to claim 3,
The statistical analysis unit calculates the appearance frequency of the face for each region on the input image by analyzing the position and size of the face detected in the past in the input image,
The monitoring system according to claim 1, wherein the selection control unit controls a size detected by the detection unit or an object to be detected according to the appearance frequency. The monitoring system according to claim 4;
An input unit that accepts a query image specification from the user,
A search unit that searches the storage unit for the object having a feature amount similar to the feature amount extracted from the query image;
A display unit for displaying a result of the search unit. The image search system according to claim 8,
The storage unit includes a feature amount storage unit that stores the feature amount, and an image storage unit that stores the input image,
The feature amount storage unit adds and accumulates information that identifies the detection method and the extraction method used in the extraction to the extracted feature amount,
The image display system, wherein the display unit displays a search result for each of the detection methods or the extraction methods.

Images