Autonomous driving is a anticipated technology in both research groups and industry. Naturalistic human driving behavior offers guidance for autonomous driving in terms of human driver reaction under certain scenario, ground truth of detection and path planning, statistics for functional safety and SOTIF, etc. How to efficiently obtain the target market naturalistic driving scenario for SAE L3 and above highly automated driving (HAD) systems is a major challenge on the road of their series production. Based on above demand, this paper uses Unmanned Aerial Vehicle (UAV) to collect traffic data from congested highways and expressways in China, conducts data preprocessing and quality evaluation, target vehicle recognition and tracking, and structured driving scenarios mining. It is expected to provide data support for HAD development in the Chinese market, including human driver behavior analysis, quantitative metric extraction, real driving test scenario generation, etc. We call this large-scale naturalistic traffic flow dataset as Aerial Dataset for China Congested Highway & Expressway (AD4CHE). It contains 5.12 hours aerial survey data of 53761 vehicles in 4 different cities in China, with a total driving mileage of 6540.7 km, 16099 lane changes and 3331 cut-ins. Some initial findings from AD4CHE are introduced. The dataset will be open sourced at https://auto.dji.com/cn.