My work is focused on analyzing and processing extremely large-scale driving datasets, developing machine learning and deep learning models to reduce dangerous driving scenarios and improve the function of autonomous driving, and deploying algorithms on vehicles, cloud platforms and servers. The projects I am responsible for include identifying lane-level road construction areas and empowering vehicles with the over-the-horizon perception of road construction areas, detecting different levels of driving accidents for cause analysis and function optimization, scoring driver behaviors when autonomous driving is running to help drivers develop better driving habits, and monitoring function abnormalities of autonomous driving and operating anomalies based on scenarios. The detailed work experience can be summarized as follows.1. Identifying Lane-level Road Construction Areas and Empowering Vehicles with Over-the-horizon Perception of Construction AreasI lead a small team of three persons to identify lane-level road construction areas by applying machine learning and computer vision approaches to the data collected from hundreds of thousands of vehicles. The information can help vehicles perceive construction areas out of sight. The method reduces hundreds of accidents related to construction areas and saves tens of millions of yuan. I proposed rules to collect data and deployed them on vehicles, developed a real-time construction area detection algorithm in the cloud, operated construction areas based on the map in the front end, and developed visualization tools to inspect, monitor and diagnose construction area detection. The decision tree and the Residual Neural Network (ResNet) were applied to improve the accuracy and reduce human labor. 2. Detecting Different Levels of Driving AccidentsVehicle accident detection consists of the accident recognition algorithm deployed on vehicles to collect data, data preprocessing, feature engineering, the collision identification algorithm, conclusion, and visualization. I employed rules, clustering algorithms, and gradient-boosted decision trees (GBDT) in vehicle accident detection. More details are omitted due to the character limtation.3. Scoring Driver Behaviors When Autonomous Driving Is RunningOmitted due to the character limtation.4. Monitoring Function Abnormalities of Autonomous Driving and Operating Anomalies Based on ScenariosOmitted due to the character limtation.