This project aims to recognize the license plates in the general situation. The robust and real-time model works well even if the environment is extremely bad such as darkness, high light intensity, low resolution and so on. The framework includes three parts: the license plate detection, image refinement, and character recognition.

In real-world situations, symmetric loss distribution assumption is often found too idealized, because pictures under various illumination and angles may suffer from multi-peaks, asymmetric and irregular noises. To address these problems, this paper assumes that the loss follows a mixture of Asymmetric Laplace distributions and proposes a robust Asymmetric Laplace Adaptive Matrix Factorization model(ALAMF) under bayesian matrix factorization framework. The experimental results demonstrate that our ALAMF model works well on Matrix Reconstruction, Text Removal, Face Reconstruction and so on.

The Internet of Things (IoT) is fundamentally changing our lives. However, little effort has been made to design a model tailored for the IoT rule recommendation. We not only need to re-define “users” and “items” in the recommendation task, but also have to consider a new type of entities, devices, and the extra information and constraints brought by them. To handle these challenges, we propose a novel efficient recommendation algorithm, which not only considers the implicit feedback of users on rules but also takes user-rule-device interactions and the match between rule device requirements and user device possessions into account. Experiments show the effectiveness and efficiency of our method.

This task aims to detect objects in the autonomous drive situation. We use the millimeter microwave Radar and 64-line LiDar to get the Pulse signal and point cloud data every 20 milliseconds, and detect the objects such as cars, trucks, pedestrian and any other potential obstacles. This work was done when I was an internship at FABU Technology.

The intrinsic parameters estimation of an image is an essential prerequisite for many computer vision tasks. Most traditional methods are based on line segments detection and EM refinements while all pipelines of the deep learning methods are not carefully designed and don’t make sense. In this task, we follow the ”manhattan world” assumption and proposed an end-to-end deep convolutional neural network named FocalNet to estimate the intrinsic parameters directly from a single image. The final experiments show that our FocalNet outperforms state-of-the-art methods in terms of the proposed metrics.