Sightcorp logo

Frequently Asked Questions

     

CrowdSight Toolkit - General Questions

can I track multiple people at once?

Yes, the Toolkit can analyse multiple faces at once. There is no maximum limit of faces detected in one frame but the performance of multi-face analysis depends on the power of the processor (e.g. the more people to analyse the more powerful processor is needed).

can Toolkit detect occluded/tilted faces?

Certain degree of occlusion and tilt is acceptable but faces in extreme angles (profile) can be difficult to analyse. The main features of the faces such as eyes, nose, mouth need to be visible for our software to be able to detect it.

do I need internet connection?

Our software runs offline and locally. The CrowdSight Toolkit requires one-time internet connection at start-up for license authentication, but it does not require internet connection for the real-time analysis. You will need an active internet connection if you wish to send statistics to CrowdStats. Learn more in CrowdStats General FAQ.

how does Toolkit licensing work?

CrowdSight Toolkit works with a time-based license. You have the option to choose the amount of time that fits your project best (minimum purchase order is 1 year license). The volume is determined by the number of cameras (1 license per 1 camera).

is there a maximum limit for analysed faces?

No, there is no limit for faces however the performance of multi-face analysis depends on the power of the processor (e.g. the more people to analyse the more powerful processor is needed).

what operating system does Toolkit run on?

Our CrowdSight Toolkit is currently available on Windows, macOS and Linux (upon request).

CrowdSight Toolkit - Technical Questions

can Toolkit analyse a pre-recorded video?

Yes, you can analyse videos from your PC with the Toolkit. Go to Input tab and select the video file you want to use under the Video File option. You can check the Loop video checkbox if you want it to keep rolling. Double check if Show Visualisation checkbox is checked (by default).

If your video file has low-resolution go to Input Settings and use Scale to improve the accuracy. If your video file has very high resolution (4K or 2K) image scaling is not advised as this can influence the speed of the analysis.

how to connect an IP camera to Toolkit?

In order to connect an IP camera to the Toolkit, you have to setup an URL for IP camera. This URL is defined as:

<protocol>://<username>:<password>@<ip_address>/<data_format>

For example:

rtsp://john:pAssw0rd@192.168.1.200/mpeg4/media.amp

or

http://john:pAssw0rd@192.168.1.200/axis-cgi/mjpg/video.cgi

The protocol and data format is specific to your IP camera and can be found in the manual.

Once you have the correct URL, enter it in the Input tab of the Toolkit. We recommend running CrowdSight Toolkit on a machine that is connected over wired LAN to the same network as the IP camera.

how to connect the Toolkit to CrowdStats?

First you need to register an account on the website of CrowdStats. Once your account is approved, you can login and add your first project. In the top left corner of your CrowdStats dashboard, you can see your Application Key. Copy this key and enter in the Output tab of the Toolkit under Application Key. Make sure to also check the option Send crowd statistics to CrowdStats. Finally, enter the name of your project in the Location label field. Then start the Toolkit and wait for the first results to appear (after 10-15 minutes).

how to improve face detection accuracy?

There are a few things you can do to maximize your detection accuracy.

First, go to General Settings and scroll min face size to the left to capture small faces and max face size to the right to capture bigger faces. If you are running on a powerful hardware, you can select the highest accuracy option (slide bar to right) in the Performance Settings.

In the Input tab you can also increase the camera resolution for instance from 640×480 (default) to 1280×720. If your camera doesn’t support higher resolutions, you can scale the input in Input Settings by moving the Scale slider to the very right.

An alternative to these solutions is to change your camera positioning (e.g. move the camera to have more frontal detections, increase/decrease the vertical camera angle such that it captures more faces).

How to increase detection distance?

If you’re interested in analysing faces further away, you can manipulate the Zoom slider under Digital Zoom in Input Settings. If you use this option, you will need to adjust the Image Position (arrows) to keep the area you are interested in within the frame. Please refer to How to increase face detection? for additional suggestions.

How to increase performance/speed of Toolkit?

The best way to achieve high performance with the Toolkit is to run the software on a powerful processor (e.g. i5 or i7). The software is running sophisticated computer vision algorithms, therefore, it requires certain levels of computing power. If upgrading your hardware is not an option, you can manipulate several settings in the Toolkit to optimize the performance and increase speed.

In the General Settings you can select highest performance (slider to the left) in the Performance Settings. Note that this setting uses our least accurate face detector, therefore, some false positives and undetected faces are to be expected. The highest accuracy option uses our most accurate face detector but also requires higher CPU consumption.

If you are only interested in a limited field of view, set your Region of Interest (ROI) in the Input Settings tab. This saves processing power and can also reduce false positives.

Uncheck the features that you’re not interested in analyzing (e.g. mood, clothing color).

For a minor speed bump, you can set the number of threads equal to the no. of cores in your machine but no more. To learn more, please refer to How to use threads?

How to start the Toolkit automatically?

The Toolkit can be started from the exe file normally. The start.bat file is generally used to start the Toolkit at system bootup in background with the specified parameters (please refer to our Toolkit documentation for more information on auto-restart). Adding the “–nohide” option at the end of the command line in the start.bat file will enable the Toolkit to be launched with the graphical interface and not in background. Learn more about command line arguments.

how to use threads?

Toolkit uses threads to carry out analysis of features such as age and gender in parallel. The number depends on the number of faces to be analysed and the number of cores in your machine. If you are planning to analyse multiple people at once, we advise setting the threads to 1 per core.

If the number of threads is set to 2 (by default) the Toolkit will analyse 2 faces simultaneously (using 2 separate cores) and then move onto the next face. For a minor speed bump you can try using threads equal to the no. of cores in your machine but not more (setting higher nr. of threads can have a reversed effect).

The maximum is 16 threads per Toolkit instance. This, however, does not mean that the Toolkit can only analyse 16 faces at once but that it can process 16 faces in parallel – given that you have 16 cores in your processor (i7 processor has between 4-6 cores).

what is the data consumption for CrowdStats?

The consumption depends on the number of people detected per second, but it varies between 500bytes to 1 kilobyte per second.

what is the required storage for the CSV file?

The size strongly depends on the number of people analyzed since the Toolkit dumps raw data without aggregation. S single file for the Toolkit processing a video of a full crowd for 30 seconds generates around 30kb of data (excluding landmarks). Interpolating the above to a full day of data, this would mean a CSV can generate around 60kb of data per minute, which is around 3.6MB per hour and around 86MB for 24 hours. That would be around 30GB of raw data in a full year.

The Toolkit does not do rotation on these files (removing old data and write new data to never exceed for example 1GB) by default so once you start writing to CSV, you need to make sure you have enough space on the disk depending on how long you want to run the software unattended.

Crowdstats - General Questions

do I need internet connection?

Yes, the CrowdStats reporting dashboard is a web-based service which requires active internet connection to process data from the Toolkit. The data is sent from Toolkit to CrowdStats every second. If there is no internet connection, the data will not be pushed to CrowdStats (only stored locally by Toolkit). The day reports in CrowdStats are updated every 10 minutes, week reports every 1 hour, month reports every 6 hours. At this moment, it is not possible to upload data from the local storage or CSV into CrowdStats.

how does licensing work?

CrowdStats works exclusively in combination with CrowdSight Toolkit. Both of these products work with a time-based license. You have the option to choose the amount of time that fits your project best (minimum purchase order is 1 year license). The volume is determined by the number of cameras (1 license per 1 camera).

what operating system does CrowdStats run on?

CrowdStats runs on any operating system since it’s a web-based service. For better compatibility, we suggest using it on Desktop devices (Windows, OSX, Linux) with Chrome browser.

Crowdstats - Technical Questions

how to connect CrowdStats with the Toolkit?

First, you need to register an account on the website of CrowdStats. Once your account is approved, you can login and add your first project. In the top left corner of your CrowdStats dashboard, you can see your Application Key. Copy this key and enter in the Output tab of the Toolkit under Application Key. Make sure to also check the option Send crowd statistics to CrowdStats. Finally, enter the name of your project in the Location label field. Then start the Toolkit and wait for the first results to appear (after 10-15 minutes).

how to rename a camera in my project?

There are two ways to change the camera name.

You can change the Location label in the Toolkit. This action will register a new source in CrowdStats when you start the analysis (new graphs) while preserving all data from the previous camera under the old name. Note that you will need an available slot on your license to add a new camera.

If you want to change the name in CrowdStats, you need to delete the camera, change the name in Toolkit and then register it again (start the analysis). The previous data from that camera will still be stored in CrowdStats but it will be marked a different colour.

what is the data consumption for CrowdStats?

The consumption depends on the number of people detected per second, but it varies between 500bytes to 1 kilobyte per second.

   

 

 

Technical Specifications

The table below shows how FaceMatch SDK performs on the Labelled Faces in the Wild (LFW) dataset:

FPRTPRThreshold (Inverse of distance)
0.10.99900 ±0.002130.55448
0.010.99667 ±0.005370.59791
0.0010.99367 ±0.006050.62989

FPR = False Positive Rate
TPR = True Positive Rate

These results are an indication only and are based on the specific dataset Labelled Faces in the Wild. Customers can expect similar performance, with possible variations due to hardware and the availability of annotated data.