Sustainable Development Goals

Abstract/Objectives

Computer vision is the key to enabling machines to perceive the world; many mobile devices (such as quadcopters, smartphones, robots, and smart glasses) use cameras to perform visual perception tasks such as object recognition and tracking. Implementing deep learning-based visual perception tasks on mobile devices significantly improves recognition accuracy, while also posing challenges for real-time operation and low-power hardware systems. The solution proposed in this project will combine engineering problem-solving with a biomimetic perspective. By analyzing and studying the visual system of fruit flies, we develop biomimetic optical flow algorithms, depth sensing, and obstacle avoidance neural network. We focus on visual sensing and chip integration, and design an intelligent vision system that is low-power, real-time, and capable of recognition and trajectory prediction.

Results/Contributions

The neuromorphic intelligent visual system chip developed in this project can not only be used for object recognition and tracking but also has the potential to be applied in the following situations: (1) In road traffic, it can be used in surveillance cameras to automatically detect violations and upload them to relevant departments. (2) In the home environment, it can assist caregivers in looking after the elderly and children and immediately notify family members in case of accidents. (3) In security system applications, it can achieve higher security through facial recognition to prevent unrelated individuals from entering. (4) In other applications, it can even be used in unmanned surveillance of dangerous areas related to national defense. Therefore, the development of the neuromorphic intelligent visual system chip in this project has broad applicability and is expected to have significant commercial value. It also has the potential to improve the quality of life and benefit a high-quality living environment.

Keywords

Neuromorphic chip obstacle detection optical flow depth estimation obstacle avoidance

References

Contact Information

羅中泉
cclo@mx.nthu.edu.tw