Event sensor is a novel sensor that offers unique advantages over common sensors. But its data is rather limited. Relying on manual data collection can become a bottleneck for algorithm development. So a more efficient approach is needed. A recent research trend is to replace real data with synthetic data, and one notable approach is Infinigen. Infinigen is a natural scene generator that generates assets procedurally for RGB data. Thus, the generation process is automatic, randomizable and controllable, allowing infinite variations of data. Inspired by Infinigen, this project aims to procedurally generate event data for built environment. Particularly, it generates event data for traffic detection. First, a procedural traffic environment, where roads and intersections are generated procedurally and resemble real world environment, will be developed. Second, we develop an algorithm that creates challenging layouts such as occlusion, rare viewpoint and crowded scenes. Third, the rendering parameters are optimized through an evolutionary algorithm to generate synthetic event data better for real-world applications. We train an algorithm on the data and evaluate it with real-world data. This has the potential to be extended to other smart city applications. The ability to generate infinite and diverse data is important to algorithm development.
