Closed gavinjalberghini closed 2 years ago
Written Abstract:
Embedded computing devices have emerged as a cost-effective way of implementing machine learning on a mobile platform for use in everyday life. This research compares multiple implementations of convolutional neural networks on the Nvidia Jetson Nano for the purpose of animal recognition. We ultimately compare the use of a YOLOv4-tiny algorithm with optimized Tensorflow implementations to both demonstrate the capability of this technology for this application as well as determine which of the two prominent approaches would work better in this scenario. Both algorithms are evaluated on a set of incoming still images for metrics as well as tested with real-time footage to determine successful recognition ability. We found that animal detection and identification is very much achievable on the NVIDIA Jetson with a satisfactory frame rate, supporting our hypothesis that there are affordable computing solutions in this domain. Though both algorithms produced viable results, we found that certain constraints change which algorithm would be desired in this context.
Due March 14th
Acceptance Criteria: