National Hotline:400-000-9680

Company news

Current Position:Home | News | Company news

Intelligent Video-based People counting camera Solution
Author:admin  Release date:2014-09-12  Hits:87
Article: Intelligent Video-based People counting camera Solution, July 2012

Media: Intelligent Building

Issue: 143

Intelligent Video-based People counting camera Solution

As Intelligent Video Surveillance (“IVS”) technology becomes more mature and advanced, video-based people counting camera has been a regular tool for statistics, analysis and mining of traffic data. Built on model-based computer vision, the system counts number of people entering/exiting accesses of passages, provides information including crowd flow direction, and features high accuracy, high adaptability and low OPEX. Its powerful central traffic analysis capability provides users with options from dozens of reports and integrates with third party software like ERP, POS or CRM to help mangers to analyze traffic by time and space efficiently and make quicker and more informed decisions.

I. Working Principle

People counting algorithm is the soul of the entire system. The system locates and detects individuals by two phases, i.e. detection and tracking to ensure counting accuracy. While people are distinguished from things at the detection phase, number of passerby during a given period of time, including identification of access direction is confirmed at the tracking phase.

1. Detection Phase

Model-based computer vision technology is used at the detection phase:

Step 1: Loading model. In the system, a model can understand a set of features of a “person” in a certain context (mainly about angle). Firstly, images of many people at a given angle in a scene are captured, which are called sample. Once adequate samples (generally about 100,000) are obtained, features are learnt through those samples to derive parameters which are referred to as model so that model can be named a set combining features of 100,000 people. Figure 1 indicates different samples in the same scene, and system detection is initialized by loading corresponding model in the case of a similar scene.

Different samples in the same scene

Step 2: extraction of features. Relevant features shall be extracted to distinguish things from people with the model. According to results of automatic machine learning, a model is composed of 90% data extracted based on shape and 10% data extracted based on color and texture. The red line shown in Figure 2 explains the feature of the shaped extracted.

Feature of people under three conditions

Step 3: Output result. In practice, a model can be deemed a filter while feature is material to be classified. Material meets criteria is released (i.e. which is considered as “people”) otherwise the material will be intercepted. People can be distinguished from things through computation based on model and feature.

2. Tracking Phase

Feature related technology is also used at the tracing phase. In addition to color and shape, features to be tracked also include location. All people in the previous frame of image are compared with those in the next image by computation and two people who have the most similar features are considered as the same person. Complete trajectory of a person can be determined by comparing all images, pedestrians passing through the detection zone may be determined by computing the number of entries of trajectory, and whether a person exits or enters the detection zone can be determined by computing the direction of trajectory. The tracking process is shown in the following two groups of figures.

Original image; detection of people; target tracking
Model-based detection mechanism can play a greater role in certain particular scenes and is particularly advantageous when camera is installed vertically rather than obliquely for example. In addition to wide access and low ceiling, camera must be installed obliquely in objective conditions where it cannot be installed vertically due to high elevation (above 10m) or special ceiling material during practical implantation of project. In such case, camera must be installed obliquely and model-based detection mechanism is more suitable, while conventional technology based on background modeling fails to deal with such circumstances.
I. Architecture
Hardware of this solution includes camera, traffic analyzer, network equipment and central management/data server. Typical architecture of the system is shown in the figure below.

Point of presence (POP); desktop client; central management server; hand-held client
The entire system can be deployed flexibly at the most suitable video POP of respective zone in conjunction with location of front-end camera POP and traffic analyzer to analyze front-end video in a real-time manner. Traffic data obtained through analysis is sent to central management/data server via network. All permitted users within LAN may check result of people counting and analysis through Web.
III. Software Function
1. People Counting
The system uses model-based computer vision technology to obtain accurate traffic data by detecting and tracking people in the scene, counts number of people exiting/entering accesses at passages, and provides information including crowd flow direction and speed. Users may designate the system to monitor one or more than one access or count traffic in one or two directions.

             Indoor vertical angle;                                         Heavy traffic;                                Outdoor inclined angle

2. Traffic Analysis and Management Platform
The traffic analysis and management platform has graphical navigation oriented display function that allows display of indices by nation, region, plaza, zone, main access, floor level, and store step by step, provides dozens of reports, and compares current traffic data with previous traffic data in each detection position of camera and each zone on a daily, weekly, monthly and yearly basis, and meets users’ needs including knowing traffic and its change during each period of time; difference of conversion ratio and traffic between business days, weekends and holidays; distribution of traffic by time and location; and comparing with historical traffic. Besides, the system provides customized reports to save data reorganization time significantly.
Multi-dimensional Analysis
Analysis report supports multiple time, spatial and logical dimensions.
A. Time dimension: 10min, hour, day, week, month, year, etc.
B. Spatial dimension: door, store, floor level, plaza, etc.
C. Logic dimension: business type, etc.

Diverse Reporting Styles
The system refines type of report and provides report in formats required by users, including:



Best Format

Formats available

Data List

Listing a series of indices of a series of plazas/stores, suitable for integrated survey against plazas/stores for a given period of time


Clustered chart

Data Table

Displaying an item of a series of plazas/stores, suitable for intuitive comparison of plazas/stores

Bar graph

Sheet, pie chart

Comprehensive Comparison Report

Displaying certain quantitative indices of a series of plazas/stores in a comparison manner, suitable for in-depth comparison of plazas/stores

Clustered chart


Trend Chart

Displaying certain quantitative indices of a series of plazas/stores in a comparison manner over time, suitable for comparison of trends

Broken line graph


Distribution List

Displaying any two quantitative indices of a series of plazas/stores in a comparison manner within a plane, suitable for data mining

Scatter diagram

Sheet, clustered chart

IV. Benefits
1. High Accuracy and Adaptability
Ø model-based computer vision technology, locating and tracking pedestrians accurately, 95% or higher accuracy;
Ø 90% or higher accuracy under heavy traffic and other extreme conditions;
Ø high anti-interference capacity, distinguishing shadow from people or things effectively, immunity to influence by change of lighting, black carpet, door curtain, rapid movement of people and other factors;
Ø filtering out people who stay still, wander around or go across the detection zone;
Ø capable of detecting traffic at an interval of more than 1cm.
2. Powerful Traffic Analysis and Management Functions
Ø powerful central data processing capacity, managing data of more than 100,000 cameras for several decades;
Ø high compatibility, standard XML protocol, supporting interconnecting with platforms of third party system;
Ø supporting multi-level networking architecture, allowing multi-user login and hierarchical authorization management;
Ø rich and diverse data display formats, intuitive and diversified reports including user-defined report;
Ø graphical navigation oriented display capability allowing display of indices by plaza, floor level, zone, main access and store step by step;
Ø providing analysis reports, including, without limitation, conversion ratio, density of traffic, rank of stores by floor level and category, and comparison over the same period and consequential comparison of plazas and stores.
V. Conclusions
People counting and analysis is a key market survey tool that provides accurate data as reference for operation decision-making and comprehensive management of large supermarkets, malls, shopping centers, chain stores and other business systems, and also provides passenger data necessary for daily management of parks, tourist attractions, museums, exhibition halls and similar fields. As the typical application of IVS in business field, video-based people counting camera has been widely used in commercial property, chain store and similar industries and is expected to expand extensively from concentrated fields to various fields with constant development of key technology.