Current trends of environmental awareness combined with a focus on personal fitness and health, motivates many people to switch from cars and public transport to micromobility solutions, namely bicycles, electric bicycles, cargo bikes, or scooters. To accommodate urban planning for these changes, cities, and communities need to know how many (micromobility) vehicles are on the road. In this work, we design a concept for a compact, mobile, and energy-efficient system to classify and count micromobility vehicles. We aim for a battery-powered system, which can be installed in a short time by a single person and is non-invasive for the ground or traffic. For this purpose, we investigate a system architecture consisting of uncooled long-wave infrared (LWIR) imaging sensors and a neuromorphic co-processor. This processor can perform classifications in a fixed time of 16,2μs without requiring additional hardware for learning. It can adapt its knowledge base to new environments without external updates. Our system is aimed to achieve adequate accuracy only using a low-resolution sensor and moderate processing power. We will show that an object representation of only 256 Bytes is sufficient to achieve a four-class classification accuracy higher than 80%, with an overall power consumption of less than 7mW. We estimate a total operating time of at least two weeks using a common 12V 17Ah lead battery.
|