There is an increasing demand for efficient driver-assistance systems in the vehicle industry. As hardwares and technology have became more advanced, neural network-based solutions in object detection get a highlighted role in this topic. We need to consider that in embedded systems the amount of available memory- and processing capacity might be limited. Because of this reason, we need to make a trade-off between the performance and the hardware-requirements of our solution.
In my current thesis I have examined the existing network-structures according to this problem, furthermore the possibilities of making an existing neural-network more efficient. There are several existing optimalization techniques that aim to reduce the memory-consumption and the number of computing operations. We can even call the intention of getting better results with a given complexity as optimalization.
I have made some literature research in connection with these topics, and I have tried to use the gained knowledge in practice.