Having looked into vision based security devices, it became evident that there was not a cost-efficient way of carrying out facial recognition based security without expensive hardware to identify potential threats and prevent people being harmed. Such a solution has many benefits ranging from use in one’s home to let them know when a loved one has returned home safely to use in schools to ensure that no unauthorised individuals are able to enter the building. This system could also be used by businesses to alert them when known shoplifters enter the premises so appropriate action can be taken. Furthermore, it could be used as a hands free entrance system to allow the elderly and disabled to gain access to their homes by simply looking into the camera.
Pre requisites:
miestenapteekki.comThe program was written in Python makes use of the tensorflow library to deal with the machine learning models of the program. Opencv is also used for the video feed and the Twillio Api is utilised to help with the sending of the text messages.
Threading had to be used as the video feed needs to be able to run in parallel with the rest of the functions of the program.
How does it work
- The administrator is expected to upload faces of people the program should detect. A model is then trained on all the faces that have been uploaded.
- A live video feed from the camera will first be analysed for faces. If a face is present, it will analyse the face against the model to try and recognise the face.
- The administrator will receive a text notification as soon as a recognition is made. Additionally, each detection is logged into a database. A video snippet of that time period is also recorded and saved.
- Administrators will have the chance to view each detection at a later date/ generate reports and search for detections.
Limitations:
One of the main limitations of the software is that the accuracy of the facial recognition will be dependent on the quality of the camera that the user decides to connect to the software as well as the quality of the images the user uploads of known faces. A lower quality camera will mean that the resolution of the frame will not be as high, as a result, the facial recognition algorithm is more likely to incorrectly recognise the face.
Another limitation will be that the software will not be able to perform facial recognition in the night. Even if an Infrared camera is used, the lack of detail will mean that the classification algorithm will not have enough data to accurately identify a face.