An effective image processing system for road accident investigation

Project title

An effective image processing system for road accident investigation

Project outline

In an incident of major collision of vehicles on roads, the virtual reconstruction of the collision scene is a very important part of investigating what has caused the incident and which party is potentially at fault. To do this, the investigation detectives are required to collect visual data at the scene of the incident using various equipment. In some of these incidents, the general public, who have witnessed the incident, might also be requested to provide data (e.g. images and videos) related to the incident, captured on their mobile devices in order to assist in the investigation.

To date, due to the differences in modality of the equipment, the formats of the visual data greatly vary. Thus, the detectives are spending enormous amount of time and effort in tediously verifying and manipulating the data before these data could be used to restructure the collision scene and useful information to aid the investigation could be extracted. It is now also difficult to present this information to the judges and juries in court to aid them in understanding the findings of the investigation and evidence.

Thus, to address the issues stated above, the aim of this research project is to investigate, design and develop an image processing system which can efficiently and accurately verify and collate visual data captured from equipment of different modalities for restructuring the virtual collision scene and for controlling the presentation of the investigation findings/evidence in a meaningful manner in court.

Research Aims

As stated above, the aim of this research project is to investigate, design and develop an image processing system which can efficiently and accurately verify and collate visual data captured from equipment of different modalities for restructuring the virtual collision scene and for controlling the presentation of the investigation findings/evidence in a meaningful manner in court.

This system comprises the following four modules:

  • a user-friendly software platform to integrate images, videos, and other forms of multimedia data of a road accident scene so that it can be used to (i) analyse the crime scene more comprehensively; (ii) improve evidence cataloguing by fusing multi-modal imagery; and (iii) produce accurate 2D images and hard copies that will be admissible in the courts.
  • integrating aerial drones to the existing scanning system for aerial scanning of crime scenes with active and automated drone manoeuvering to optimally cover occluded regions and improve scanning precision in regions of greater significance e.g., related to evidence.
  • an image/video filtering module which can automatically classify images/videos (e.g. captured on vehicle dash cams and mobile phones of passengers), which are submitted by general public, based on their relevance to the road crime scene. This module will alleviate detectives from manually filtering out relevant images/videos to be potentially used as evidence in the investigation.
  • an Augmented Reality module to describe the crime scene and present evidence to the members of the juries.

Project supervision team

Shyh Wei Teng, Manzur Murshed and Adrian Shatte (ECR)