Deep learning excels in recognizing objects in images as its implementation using 3 or more layers of artificial neural networks where each layer is responsible for extracting one or more feature of the image. And Neural Network is a computational model that works in a similar way to the neurons in the human brain. Each neuron takes an input, performs some operations then passes the output to the following neuron. By combining both, we are able to isolate the facial characteristic into 512 dimension hyperspace and use that into facial recognition to ensure the accuracy.

GPU-accelerated computing is the use of a graphics processing unit (GPU) together with a CPU to accelerate deep learning, analytics, and engineering applications. Our software system allow GPU to participate in image processing and neural network computing process, which largely improved the speed of detection and analysis. GPUs have thousands of cores to process parallel workloads efficiently. Our implemented facial recognition speed is up tp 1000 fps.

The detection and analysis process starts from detecting faces with a pre-trained models. The face then transformed and corpped for the deep neural network. This transformation process will try to make the eyes and bottom lip in the same location on each image. After that, the system use a deep neural network to represent the face on a 512-dimensional unit hyperspace. The step gives a generic representation for anybody's face. Unlike other face recongnition, it provides the nice property with a larger distance between two face, which means that the faces are likely not of the same person. This property makes clustering, similarity detection, and classification tasks easier than other face recognition techniques where the Euclidean distance between features is not meaningful.

To enhance the security of our system, both IR technologies and dual-camera hardware are adopted for false detection. IR is mainly used for non-life objects detection, like face mask. The dual camera can capture two different position faces in the same time for data screening and another method for preventing face picture detection. With both of them, face mask or any non-human objects will not be able to pass through our security system.

Data analysis to target specific group of customer has been a very important strategy for most business in nowadays. However, it is awkward to collecting those information via survey and may have negative impact on business. Our deep learning based cloud analysis software on Facial Age Estimator comes to handy in this situation to help business owner to distinguish the customer groups. This project is based on deep learning technologies (AAM,LBP, GW and LPQ) and cloud service frameworks. Beside the collection of non-personal identification information, we will also provide business analysis service for clients to adjust their products for more targeted user.

Human gait is a new biometric aimed to recognize individuals by the way they walk , which has become more vital feature in nowadays smart home system. In this project, we are proposing a new novel method to develop the gait recognition, and integrate with facial recognition system to improve the robustness. In this way, we will not be able to identify the persons with or without access in distance, but also give more detail recognition information for further analysis.

Deep learning based computer vision techniques can be applied at public parking lot to monitor the traffic coming in and going out. The benefit of such system includes automatically monitoring the available slots in the lot, timing and calculating the parking fee by creating a profile for each vehicle. Others benefit includes security control, thus, such system can also be applied in closed community. Hence, we propose building a access control system that can be applied at parking lots and closed communities based on the car model and license plate extraction and recognition using Deep Neural Network (DNN).

Pet doors make pet owner life easier, however, the biggest problem with regular pet doors is other animals may get into owner home. Therefore, we propose to develop a smart pet door control system which only allows the recognized pet inand out. This system is based on the deep neural network system, will have more accuracy and modes to pet access control. Similar to human facial recognition system, each pet has their own facial characters that can be used to recognize them. Comparing to other smart pet controller system like key fobs, the owner does not have to worry pets lose their access due to losing the key fobs.