A client sought to build a SaaS (Software-as-a-Service) platform for Digital Evidence Management (DEM) to support a service for the collection, aggregation and management of digital evidence intended for law enforcement agencies. This involves audio & video clipping, computer vision and image & video processing functions.
Alumnus built the media processing core of the DEM to perform the clipping, computer vision and processing functions. An in-house library was developed to natively split video and audio streams of a wide variety of media formats, without re-encoding or transcoding –preventing quality loss and preserving authenticity.
Image quality was improved using custom libraries developed for proprietary algorithms. This was used to identify faces using Deep Neural Network (DNN), and license plate and street names using Optical Character Recognition (OCR). The algorithm was further improved to clean false detections in both video and still media, and track objects in videos while accounting for abnormalities like unpredictable movements and occlusions. Alumnus’ solution also allowed to add annotations, selective pixel-level adjustments and redaction to protect privacy.
The solution could track up to 10 objects per frame with random movements and occlusions; which increased to 100 objects per frame with further enhancements.
Technology Stack:
OpenCV, Object Detection, Deep Neural Network (DNN), Optical Character Recognition (OCR), Image and video processing, pixel adjustments, C++, C#
