Video sensor digital forensics have become necessary since the start of closed-circuit television (CCTV) cameras. With the emergence of CCTV, crime detection became less time-consuming, as CCTVs have been installed everywhere from homes to public places ^[1][2][20,21]. CCTVs have been considered a great strategy for analytical improvement in crime detection, as they enable temporal and spatial analysis ^[3][4][5][22,23,24]. The United Kingdom CCTV industry showed that with the installation of CCTV systems, the crime rate decreased significantly ^[6][7][25,26]. CCTVs require video processing techniques for the analysis and identification of crimes ^[8][27].

In the vehicular context, different sensors are crucial sources of digital evidence ^[9][28]. One of these sensors is the dashcam, which has been used for multiple purposes ^{[10][11][12][13]}[29,30,31,32]. Dashboard cameras have been used to increase road safety ^[14][33]. These cameras are able to record both audio and video and are mounted inside vehicles. Most of the cameras are facing forward, but some of them are dual-lens cameras that can provide footage of what is happening in front of and behind the vehicle while driving. Dashboard cameras are either mounted on the windshields or in front of the vehicle ^[15][34].

Nowadays, dashboard cameras are integrated with built-In GPS, which can provide speed, time, location and route data ^[16][35]. The dashcam industry is evolving rapidly, as the number of vehicles with installed dashcams and the expected market size will reach $5.2 billion by 2026 ^[17][36]. Dashcams provide a wide variety of features that are available and are associated with the recording purpose, including, but not limited to, GPS location of the recorded video footage, car speed during video footage, connection to personal assistance like Amazon Alexa, mobile hotspots, voice commands, emergency events, accident detection, parking mode, G-sensing, and auto-sync with the cloud ^[18][19][37,38]. Dashcams can be controlled through the camera itself if the dashcam has a screen, and/or a mobile application can be installed on a mobile device to configure the dashcam.

Dashcams are the most useful resource for vehicular crime events and accidents as the recorded footage not only provides quantitative data of car movement but also provides additional information like the driver’s vision condition ^[20][39]. The footage from dashboard cameras is stored in the memory, and this footage can be used by courts for testimony to remove uncertainties and by insurance companies. Video recording from an installed camera in a vehicle helps insurance companies deal more quickly with damages issues ^[21][40]. Dashcams gain huge significance as they help forensic experts in reconstructing road crashes and preventing distortion of the facts, which could lead to a possible miscarriage of justice ^[22][23][41,42]. A study shows how a dashcam was proved to be great evidence for putting on trial a ‘crash for cash’ fraudster who intentionally caused a car accident just to claim the insurance money ^[24][43]. Also, video evidence can clarify many cost claims by providing important details about the incident: for example, third-party pedestrian claim costs drop by 80% for vehicles with dashcams installed ^[25][44].

Ensuring the authenticity of the evidence retrieved from dashcams is important. It needs to be evidence generated by the dashcam itself and not modified by any external party ^[26][45]. This requires all available artifacts extracted from a dashcam to be connected together. For example, a fraudster can submit a video recorded by another dashcam, causing the case of a staged accident. When performing a digital investigation for the dashcam installed on the car that the fraudster claims is recorded on the video, other artifacts related to the video need to be consistent. The GPS data should point to the location of the accident, and the logs of the dashcam should indicate an emergency event that happened. When there is inconsistency between the submitted video evidence against other artifacts related to the video recording, this can be considered as fraud.

Few studies have discussed dashcam digital forensic analysis. Ref. ^[27][4] described the artifacts extracted from dashcam SD cards related to the recording mode of the camera, the GPS data associated with the recorded videos, data related to the vehicle speed and license plates that were captured by the camera. Although ^[27][4] could find a number of artifacts from the SD card of the cameras, the research did not include artifacts from the dashcams themselves or explore the logs written by the dashcams and/or the mobile applications associated with the dashcams. Additionally, refs. ^[27][28][4,5] stated the need for dashcam forensic methods that use OCR to extract metadata, microphones to record surrounding sounds, and associations between temporal and geospatial data.

The authors in ^[26][45] proposed an algorithm to extract features from the videos that are uploaded online into a publicly accessible website. The feature extraction relies on recognizing the motion blur in the video recording that is generated by the unique movement of the vehicle while the dashcam is recording. This helps digital forensic investigators identify the source vehicle from which a certain video is recorded in case that video is used as evidence in an insurance claim or as evidence in a court. Further, an open-source framework was proposed to identify a variety of metadata related to 14 types of dashcams associated with a database to store the metadata ^[29][46]. Although this researchtudy proposed systems that are able to recognize and store a lot of metadata related to different dashcams, it was not able to recognize the dashcam from which the digital evidence is acquired. Also, the authors do not support review of logs, which makes the investigations more difficult. According to ^[26][45], digital investigations need an algorithm that helps to prove if the submitted video recording is generated by the same car that had the accident.

Recently, online submission of digital forensic evidence has emerged to enable rapid investigations. This is also applied to dashcam forensics, as witnesses can upload evidence videos to portals through which law enforcement agencies can access these videos. Although online submission of digital evidence facilitates digital investigations, the increase in the number of submissions requires more processing time and memory ^[27][4]. Also, online submissions pose serious challenges to investigations, as digital evidence can be modified ^[30][47], shared with third parties ^[31][48], superannuated and manipulated ^[32][49]. Due to these reasons, this research uses offline investigations by direct reading of the dashcam SD cards.

In summary, dashcams are becoming more widespread than before, and they contain important evidence regrading vehicle incidents. Dashcam videos include several artifacts that assist digital forensic investigations. These artifacts require extraction methods that are accurate and that ensure integrity and authenticity. The most important sources of artifacts in dashcam videos are the text that expresses GPS coordinates, speed, time and date, and the speech that expresses warnings about lane departures and collision possibilities.

Dashcam digital forensic investigation requires extracting text from videos and analyzing speech. In relation to text extraction from video evidence, the authors of ^[33][50] developed a cloud-based OCR to extract text that is necessary for digital forensics. This method depends on online processing, which poses security threats to the analysis ^[32][49]. Ref. ^[34][51] presented the challenges in license plate recognition using OCR. Internal challenges are related to the software used for processing and the camera hardware and include resolution, view angles, focus speed, the processor and internal memory. External challenges are related to plate variations (font, color and plate position), environment factors (lighting conditions and surrounding effects) and camera mounting variations (inclination and distance from the plate). Also, ref. ^[35][52] used OCR to extract timestamps from videos to sort them according to time for digital investigations. However, the type of OCR used for video forensic analysis is not specified well in these studies. The research in ^[36][53] shows that OCR methods on images requires much computational time, and ^[37][54] shows that OCR methods can achieve high accuracy in different weather conditions if they are integrated with deep learning.

In relation to dashcam speech forensic analysis, ref. ^[28][5] states that the recorded speech on the dashcam about lane departure warnings and collision warnings is still analyzed manually. There is a need for digital tools that help investigators to analyze the speech and generate the forensic report. In other digital forensic domains, voices and sounds are considered important because they contain relevant evidence for the investigation. So automatic speech recognition that transcribes speech-to-text is necessary for digital investigations because text is predominant in digital investigations and the most acceptable type of evidence in courts ^[38][18]. Also, text can be parsed, indexed and searched easily, while integrating voice with digital forensic examinations is difficult ^[38][18]. The process of speech conversion to text is composed of speech detection and speech transcription. Although a number of libraries are available for speech detection and transcription, few are offline and open-source and support digital forensics without referring to third-party services in the cloud. Digital forensics require that the speech is not modified. For ensuring audio integrity, the authors in ^[39][55] propose a method that uses hash values to verify if audio has been modified or not.

In summary, the aforementioned studies did not sufficiently describe the methods used for text or speech extraction. This research agrees with ^[27][28][4,5], which states that there is a need for identifying the best text recognition methods and speech-to-text conversion methods that can be used to extract forensics metadata from dashcam videos. To address the problems in these studies, this research proposes a solution to investigate the storage media acquired from dashcams. The proposed solution is a dashcam forensics framework that is able to extract artifacts from given dashcam evidence and enable investigators to associate the dashcam and the extracted metadata with a map location that supports evidence tracking. This research examines different text and speech extraction methods, compares them, and identifies the one that can achieve better results than the others.