The University of Auckland

Project #23: Can Resource Constrained Embedded Platforms achieve real-time pedestrian detection?



Over recent years, the popularity of autonomous vehicles and computer vision has been rapidly increasing. These two systems are coming to a convergence point resulting in a steady rise in demand for these two technologies working in parallel. It is important that we maximise the potential of computer vision when providing information to autonomous vehicles to ensure they act in a safe manner, especially regarding humans in their path. Currently, human detection systems for autonomous vehicles are limited by the design and reliability of the cameras and proximity sensors.

To improve the safety of the public, specifically at heavy pedestrian and traffic intersections, we are proposing researching and prototyping a distributed human detection camera system to improve pedestrian safety. This system would function through the pairing of a distributed camera system for pedestrian detection, with an autonomous vehicle being provided real-time information regarding the humans in its vicinity. 

This project would include multiple key research components, specifically, identifying the methods of human detection currently utilised in similar systems and their viability for our proposed system. The current technologies available for such a time-critical system would also need to be analysed to develop a low-latency, reliable system.

The first stage of developing such a system would consist of an autonomous vehicle relying solely on the data collated by this external camera system to detect humans. The autonomous vehicle would not be fitted with a detection camera, allowing for isolation of the reliability of the distributed system alone. This would also indicate key parameters surrounding the calculation time of the camera system and the response time of the autonomous vehicle.

As autonomous vehicles grow in popularity and their numbers begin to increase on the roads, the need for a system like this will grow. This system could be applied in many busy regions, but a key application could be autonomous taxis executing pickups and drop-offs at busy airports and central city locations. Busy city intersections that experience high car throughput along with pedestrians would be another location in which this system could improve human safety. 


Research Components

·       Investigation into the pre-existing approaches for human detection and their relevance to our proposed system

·       Investigation into the current technologies i.e MCU, FPGA that could be used for this system and their relevant advantages for time-critical systems

·       Prototyping a camera network that will collate data on object/human detection (localised to a room to begin)

·       Prototyping a network system to achieve high data throughput and low latency and handle communication of the detection data

·       Prototyping a small RC vehicle to communicate with the network









Lab allocations have not been finalised