University of Twente Student Theses
As of Friday, 8 August 2025, the current Student Theses repository is no longer available for thesis uploads. A new Student Theses repository will be available starting Friday, 15 August 2025.
Embedded Person Detection on an STM32F7 with Dynamic Vision Sensor
Gatt, Pierluigi (2025) Embedded Person Detection on an STM32F7 with Dynamic Vision Sensor.
|
PDF
2MB |
| Abstract: | Event-based vision sensors offer an alternative to RGB cameras by capturing the changes in a scene. This thesis investigates the optimisation of a convolutional neural network (CNN) for real-time person detection on an STM32F7 microcontroller with a neuromorphic vision sensor, specifically Prophessee's GenX320. Neuromorphic vision sensors offer advantages such as low latency, low power consumption, and the preservation of the recorded's anonymity. However, they pose unique challenges for implementation with a CNN due to their event-based data rather than traditional frame-based outputs. The STM32F7 microcontroller also introduces limitations for CNN deployment, such as its limited internal RAM and Flash and a lack of dedicated hardware acceleration for CNN inference. This thesis reviews and implements various optimisation techniques, such as post-training quantisation, architectural simplification and input size reduction. These efforts aim to minimise resource usage while maintaining detection accuracy. Preexisting CNN models for person detection on RGB datasets like COCO were adapted to fit on the STM32F746G-DISCO's limited 1024 KB and 320 KB of internal RAM and Flash, respectively, alongside processes dedicated to the sensor and display. These optimised models were then trained for inference with a neuromorphic vision sensor using the PEDRo dataset. The applied improvements resulted in a model with an inference time of 422 ms and an AP of 75$\%$ (@0.5 IoU on the PEDRo dataset). This study shows that object detection using a neuromorphic vision sensor is feasible on ultra-low-power hardware without significantly compromising accuracy. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Electrical Engineering BSc (56953) |
| Link to this item: | https://purl.utwente.nl/essays/107638 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page