| 總結: | Face detection and recognition is currently a hot area of research that makes use of the fundamental knowledge from computer vision, image processing and pattern recognition. In the market, there is an increasing number of devices, even wearables are installed with video capabilities. In this project, I am tasked to implement a face detection and recognition system based on video in real-time. The faces detected in real time are compared with the model in the face dataset. However, the per-frame face recognition system is limited by various factors that could affect the accuracy in recognition due to wide variability in real life environment, e.g. low lighting, different posture or low-resolution images. In this paper, I will focus on improving the face recognition performance by implementing logics to support Multiple Frame Detection and achieve better accuracy. Video clips of objects walking and still objects at fixed distances will be considered in this project. Multiple deciding factors on top of the confidence levels will be used in the recognition process. This includes having a variable that adjusts the threshold based on the size of the object’s face while walking, measuring the distance of the object face’s position across multiple frames, and mechanism to prevent multiple people to be labelled as the same person at one time.
|
|---|