Intelligent vision analysis in the field of video surveillance technology

In recent years, video surveillance has been widely used in various places. Its infiltration into all walks of life has opened up an unprecedented new situation in terms of security prevention, information acquisition and command and dispatch. However, traditional video surveillance is usually based on human-based visual monitoring and recording of video information by the system. Due to the limited human energy and the fact that each monitor may need to display multiple camera screens in turn, the monitoring personnel may be tired of the eyes and miss the scenes of the screen when they observe a large number of monitors for a long time, and the monitor cannot be extracted timely and reliably. The effective information, so even if monitored by manpower, it cannot guarantee accurate and real-time discovery problems. Furthermore, the original monitoring video is usually called after the occurrence of other events for viewing, and is used for forensics or other subsequent processing.
The specific functional requirements are analyzed as follows:

Crowd gathering

Locations such as shopping malls, theaters, and stations are prone to heavy crowding in local areas during certain hours. In order to avoid the trampling and crushing caused by crowding and illegal criminals taking the opportunity to disrupt the order and conduct illegal acts, it is necessary to carry out automatic alarm processing on the situation where too many people are gathered, and then to promptly guide and avoid dangerous events.

2. Statistics of people

The people counting system is an intelligent system that monitors the flow of people and provides accurate statistics on crowd flow and crowd flow and analysis of data. Users can set up one or more entrances and exits according to the practical situation, and can also set up a single-direction or two-way crowd flow, and calculate the total number of people entering and leaving the building, the number of people entering and leaving the building per unit time, and real-time calculation of buildings. Information such as the number of people in the internal current time is used to make statistics. Suitable for supermarkets, department stores, restaurants, schools, libraries, sports fields, tourist attractions, public transportation, train stations and airports and other densely populated public places.

3. Personnel

Some criminals often repeatedly examine the surroundings of their crime sites before committing crimes. Therefore, they will continue to work in certain areas. In order to detect and deter its criminal behavior in a timely manner, it is necessary to make an automatic alarm on the behavior of personnel. This technique is suitable for areas such as military areas, prisons and other martial law areas, banks, warehouses, museums, jewelry stores and other areas of valuables.

4. Items left behind

As terrorist attacks increasingly threaten social stability and human life safety, counter-terrorism has also become an important issue in maintaining social security and order. Terrorists often leave explosive objects in densely populated areas, causing serious casualties. The monitoring of the leftovers in public places can effectively prevent the occurrence of such incidents. In addition, when the vehicle is illegally parked, it can also be handled as an item leaving an alarm.

5. Illegal stay

In places such as subways during rush hour and rush hours, there are many personnel and large people flow, and it is easy to cause congestion in narrower areas such as aisles and stairways. At this time, if someone stagnates in a narrow area, it will affect the passage of other people, which in turn will lead to congestion and crowding. Similar situations can occur on roads with many vehicles. When an accident such as anchoring or scraping occurs on a road, it can easily cause traffic jams. If you can quickly and accurately find a target for staying, notify the relevant personnel in a timely manner. Diversion can greatly ease congestion.

6. Suddenly fell to the ground

Suddenly falling to the ground means that the human body is turned into a lying posture under the condition of erection. This function is often used in public places to prevent people from suffering from sudden illnesses or severe injuries. They have not been discovered in time and have been delayed for treatment. In addition, this function can also be applied to clinical monitoring in the medical system to realize automatic monitoring of part of the patient's state.

The above content describes the functional application of the intelligent monitoring system, which not only enables the video surveillance to perform uninterrupted and reliable monitoring 24 hours a day, but also greatly improves the alarm accuracy and response speed, and overcomes many shortcomings of human monitoring.

Intelligent vision technology analysis

Through the above analysis of the application requirements, from the technical aspects, personnel paralysis, illegal detention, artifacts left behind, illegal invasion, etc. can be classified into one category - visual behavior analysis. Behavior analysis must be based on the correct detection of the target in the image sequence and tracking, from the image sequence to extract the behavior characteristics of the moving target, and then through effective algorithms to determine and analyze the target's behavior, so the behavior analysis belongs to the level of image semantic understanding jobs. The important question of behavior analysis is how to establish the behavior pattern from the sample sequence and how to correctly classify the test sequence into the corresponding behavior pattern.

However, for different application requirements, the specific algorithm is also different. The specific analysis is as follows:

1) Personnel

Accurate defamation is a process that involves complex actions and is difficult to identify. However, indirect judgment can be made by determining the distance, time and direction of the target movement within the preset area. From the above target detection and tracking, it can be seen that on the basis of reliable tracking of the target, the coordinate distance between the frame and the frame is accumulated to obtain the trajectory length in the entire motion, so that a simpler judgment method can be used. Recognition is performed, but this method has poor robustness and is likely to cause misjudgment, which can meet the lower security level. When the security level requirement is high, the training and learning can be performed and the model can be built. The model can be identified based on the state space model. The false positive rate can be reduced. However, the number and diversity of training samples are more demanding. High, the training process is complicated.

2) Goods left over

The item left behind generally refers to the person carrying the item into the target area, and then leaving the person and leaving the item in the target area. In this case, the moving target entering the target area may be tracked. When the time when the target remains stationary exceeds the preset threshold t0, it is determined as a legacy. Due to the wide variety of legacy items, a large number of sample libraries need to be established. In addition, when the legacy object suddenly appears, there is no legacy process (the leftover process is obscured and invisible), and when the legacy object is continuously blocked by a large area of ​​the moving target, the above algorithm is not applicable, so different algorithms are required for different application scenarios. Meet the needs.

3) Illegal stay

Illegal detention involves real-time calculation of multi-target tracking and target speed. It is necessary to track and record the moving speed of each target in the scene in real time. When the movement speed v of a single target is far less than the movement speed v0 of the entire crowd, an alarm is triggered. The whole group's movement speed v0 is obtained by statistics on the movement speed of all single targets. The speed limit v0 can also be set as a constant according to the situation, which can be set independently by the user's needs. In addition, the limit value v1 can be set for the motion speed of the entire target group. When the motion speed of the entire target group is lower than the limit value, it indicates that the area has a large area of ​​stagnation, and an alarm signal that is different from the individual target can be sent to remind the work. The personnel conducted timely guidance on this area.

4) suddenly fell to the ground

The above three types of human behavior analysis generally belong to the recognition of simple human activities, and suddenly fell to the category of 'action'. For the fall recognition in simple scenes, a method based on the state space model is usually used. First, the model is established, the initial model is determined, then the training sample set trains the reliable model parameters, and finally the probability of the test sample for each known training model is calculated. The behavior pattern category corresponding to the maximum probability is selected as the category to which the test sample belongs. However, in public places where people are denser, their falling process is often blocked by the passing population. This increases the difficulty of the recognition algorithm and requires an algorithm that is more robust against occlusion to meet the scene requirements.

The following two algorithms do not belong to human behavior recognition, but they are also important applications in intelligent monitoring:

5) crowd gathering

There are generally two types of detection methods for crowd aggregation: population density estimation methods and methods based on individual characteristics. The former is based on the relationship between the overall characteristics of the crowd and the density of people gathered. Such methods are suitable for situations where people are dense in large scenes; while the latter is based on the number of people in a person's characteristics, which is applicable to comparisons. Small scenes, fewer personnel, and less severe occlusion, are more accurate than population density estimates.

The crowd density estimation method can be achieved by extracting different feature vectors, for example, the total number of pixels in the foreground population or the total number of pixels in the edge, but the most widely used is the extraction of texture features. According to the different texture patterns of populations with different densities, the population density can be estimated by establishing the corresponding relationship between population density and texture patterns, and then the population concentration can be determined. The method based on individual characteristics first detects single-person targets. Since the head and shoulder parts of the human body are the least likely to be obstructed and the human body features are more obvious, head-and-shoulder detection is more widely used. In addition, there are more methods for using the transform domain; then the statistical target The total number n of persons detected in the scene is set to the threshold value n0. When n ≥ n0, it is judged to be clustering.

6) People statistics

From the point of view of the applicable environment, the demographic statistics can be divided into two categories: one is regional statistics applicable to open scenes or outdoor environments, and the other is cross-sectional statistics applicable to passages or entrances and exits. The former generally counts the total number of people in the target area. The latter generally places the camera at the top of the channel or entrance and exits, and shoots vertically downwards to count the number of people entering and exiting in both directions. The implementation of the former is similar to the crowd gathering and will not be repeated here. The latter according to its special constraints, the camera usually captures the head and shoulders of the person, so after the detection and tracking of the head and shoulders, the target can be counted. However, when multiple people pass through the entrances and exits at the same time, because the target distance is too close, it is easy to cause the goal to merge and separate, that is, the goal of multiple people is treated as a single person, or the goal of so-called 'single person' is separated into multiple persons. There are several ways to solve this problem, in which the head and shoulder features can be replaced with head features. When the target 'passes by', the head area will not usually be merged into one target. In addition, counting for two directions requires the feature of increasing the direction of motion during the tracking process.

In addition to the above applications, there are automatic alarm functions such as illegal intrusion, fights and public property destruction. The main technical modules are similar, but their specific algorithms are different for specific application environments and functions. With the continuous development of artificial intelligence, pattern recognition, image engineering, and artificial neural networks, the application of intelligent visual analysis in the field of video surveillance will gradually replace manpower, making the development of intelligent video surveillance more extensive and complete.

Flip Clocks

Flip Clock Series, fully designed and developed by Mid-Clock R&D team, CE and RoHS Marking approved.

Flip Clock can flip down the card automatically, with 48 pieces of cards on the left and 60 pieces of cards on the right. Both AM/PM and 24 hours display are available. It depends on what we print on the card.

Since the time Flip Clock was developed, it has been one of the best-selling products till today. The way it works is mysterious. It can always inspire you when you are staring at it. At least, the Flipping can tell you that time is Flying.

We have extended different series for Flip Clock, such as Classic Retro, Box Flip, Building Series, Funny Stuff. Different series have different sense of feelings for different occasions.

We can provide customized service with ordering a MOQ at 300PCS, not only printing the LOGO on it but also customize the Font/Colors/Content of the cards, even the pictures, as long as you like, can be printed on the cards. The color of the metal frame can be customized too.

Flip Clock Factory in China, China Flip Clock Factory. Flip Number Clock Factory, Flip Down Clock factory

Guangzhou Huan Yu Clocking Technologies Co., Ltd. , https://www.mk-times.com