Real Time Human Detection Counting and Face Recognition
Real time automatic counting of people and face recognition has wide application in intelligent public transportation systems, security, surveillance logistics and resource managements. The primary challenge in developing a real-time automated face detection, face recognition system, counting will b
2025-06-28 16:28:54 - Adil Khan
Real Time Human Detection Counting and Face Recognition
Project Area of Specialization Artificial IntelligenceProject SummaryReal time automatic counting of people and face recognition has wide application in intelligent public transportation systems, security, surveillance logistics and resource managements. The primary challenge in developing a real-time automated face detection, face recognition system, counting will be system performance and speed optimization[1]. There are so many tasks and processing in every seconds of live video input. Therefore, the development of real-time face detection and recognition becomes a popular study. With increasing terrorist activities augmented demand for video surveillance it was the need of an hour to came up with an efficient and fast detection, counting and recognition algorithms. Depending on few conditions such as facial expressions, head, emotions, pose and light effect. There is a need of efficient techniques for achieving this goal[2].
When we detect human counting and facial recognition by using different strategies like Open CV (open source computer vision library) a strong library for machine learning or image processing technique. Dilated Convolutional Neural Network (DCNN) is the deep learning algorithm by which we can count the large crowd gathering of hundreds of people deals with the complexity of partially visible head of a person in the real time head count of a video feed mechanism. The technique is based on the Neural Network architecture. Convolutional Neural Network (CNN) is a deep learning algorithm most commonly recommended for applications using images because it perform the combine task for feature extraction and classification[3][4].
This work uses selected facial features and a popular multilayer feedforward neural network for the task of classification. The extracted features are determined and presented as a pattern vector to the neural network. The learning algorithm recognizes people's faces by learning the approximation of facial features, regardless of different facial movements. The feature matrix changes depending on the face motion[5].
Project Objectives- To count the number of people at any spot.
- To recognize human face for identification purposes.
- To identify a collection of data with the same face in a database of training photos.
- To count the number of individuals in or out a reference line.
- To estimate the no of people present in the shoping mall,stadium or any building by camer the no of people enters or leave from gates.To estimate the known people in the building by using face recognition techniques.
- In dense gathering areas i-e shopping malls, cinemas etc. when any natural disaster or any emergency like fire or any bomb blasting occurs people start running to save their lives. In this situation, some people stuck inside while some came outside for the sake to save their lives. To estimate the number of people that are in danger we can use this human counting system at the leaving or entrance to know how many people are in so to rescue them. Furthermore, the face recognition system also recognize those people whose data is registered in system or the members and staff of that organization in order to rescue them easily.
We will use both OpenCV and dlib to build our people counter. For typical computer vision/image processing operations, we will utilize OpenCV, and for people counting, we will use the deep learning object detector. Then we will utilize dlib to construct correlation filters. We could have used OpenCV instead, but the dlib object tracking implementation was a little simpler to work with for our project.
Highly accurate object trackers will integrate the concepts of object detection and tracking into a single algorithm, which will often be separated into two phases:
- Detecting
- Tracking
- For face detection, we are using deep neural network (DNN) and face recognition model. The face recognition model detects only the faces not any other parts of the image. For Example if we put mask or hand on our face it will not detect it as a face.
- For classification, we are using K-nearest neighbor (KNN) model.
- In our project, we give 100 images of faces for a single person the KNN model stores these images in a list and process on them. Note that we are also assigning specific name and ID to specify that person. When we test an image through a webcam, the KNN classifies with the data of images stored as a list. If it matches with faces of data set then it shows the specific name and ID of the testing image. We are also counting the known persons and storing these results in an excel file with the date.
There is no need extra hardware only two cameras one for Counting and another for face recognition.
There is the need of a computer included GPU for processing.
All the malls,stadium and other buildings have the computer with these specifications but we need a project to run and the result will be displayed.
Technical Details of Final Deliverableproject is in complete running form which counts the no people and recognize as well.
The accuracy of dense crowd is efficient and recognition efficiency is 99.99%.
During collecting data for face recognition if the face is not clear the image skips if two person are in the image the image also skips
The final results shows even more than one persons in the front of camera it recognize efficiently.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 49100 | |||
| 480 Mega Pixel WebCam | Equipment | 1 | 1600 | 1600 |
| 720 mp WebCam | Equipment | 1 | 4000 | 4000 |
| Navidea GPU 2nd Hand | Equipment | 1 | 40000 | 40000 |
| Tripod stand for overhead camera | Equipment | 1 | 1500 | 1500 |
| door of wood for recognition and other needs | Miscellaneous | 1 | 2000 | 2000 |