Face on Edge-Realtime Face Recognition
Description
Face recognition is a method of identifying the name/ID of the faces present in an image or video. This method learns the features of individual faces and differentiates the features from each other. This in turn helps in identifying unique features for each face/class. Facial recognition is one of the top emerging technologies during 2020-25.
The facial recognition system (POC) developed by eInfochips focuses on the applications rely on edge devices, which captures face close to the camera (2 to 3 feet distance), with clear face visibility, without occlusion, and with a minimum image resolution of 720 x 640. Deploying the face recognition model in an edge device is challenging than cloud based servers. Since the computational capacity is limited on edge devices, optimizing and fine-tuning the model without degrading its accuracy and performance (Frame rate) is essential. In the present work, the model is optimized using the TensorRT module and deployed on the Nvidia Jetson Xavier AGX board. The developed system is well suited in applications like home security, school/office attendance, driver verification, smart advertising in retails, etc.
Highlights
- Introduction
- Facial recognition pipeline
- Face detection
- Feature extraction
- Feature classification
- Datasets
- Accuracy obtained on the dataset
- Optimization technique
- Results
- One shot learning
- Limitations
- Conclusions
To read more, download the copy 
To read more, download the copy
To download this resource
Fill in the details below
Description
Face recognition is a method of identifying the name/ID of the faces present in an image or video. This method learns the features of individual faces and differentiates the features from each other. This in turn helps in identifying unique features for each face/class. Facial recognition is one of the top emerging technologies during 2020-25.
The facial recognition system (POC) developed by eInfochips focuses on the applications rely on edge devices, which captures face close to the camera (2 to 3 feet distance), with clear face visibility, without occlusion, and with a minimum image resolution of 720 x 640. Deploying the face recognition model in an edge device is challenging than cloud based servers. Since the computational capacity is limited on edge devices, optimizing and fine-tuning the model without degrading its accuracy and performance (Frame rate) is essential. In the present work, the model is optimized using the TensorRT module and deployed on the Nvidia Jetson Xavier AGX board. The developed system is well suited in applications like home security, school/office attendance, driver verification, smart advertising in retails, etc.
Highlights
- Introduction
- Facial recognition pipeline
- Face detection
- Feature extraction
- Feature classification
- Datasets
- Accuracy obtained on the dataset
- Optimization technique
- Results
- One shot learning
- Limitations
- Conclusions
To download this resource
Fill in the details below
Description
Face recognition is a method of identifying the name/ID of the faces present in an image or video. This method learns the features of individual faces and differentiates the features from each other. This in turn helps in identifying unique features for each face/class. Facial recognition is one of the top emerging technologies during 2020-25.
The facial recognition system (POC) developed by eInfochips focuses on the applications rely on edge devices, which captures face close to the camera (2 to 3 feet distance), with clear face visibility, without occlusion, and with a minimum image resolution of 720 x 640. Deploying the face recognition model in an edge device is challenging than cloud based servers. Since the computational capacity is limited on edge devices, optimizing and fine-tuning the model without degrading its accuracy and performance (Frame rate) is essential. In the present work, the model is optimized using the TensorRT module and deployed on the Nvidia Jetson Xavier AGX board. The developed system is well suited in applications like home security, school/office attendance, driver verification, smart advertising in retails, etc.
Highlights
- Introduction
- Facial recognition pipeline
- Face detection
- Feature extraction
- Feature classification
- Datasets
- Accuracy obtained on the dataset
- Optimization technique
- Results
- One shot learning
- Limitations
- Conclusions
To download this resource
Fill in the details below