Making Police Body Cams Work on a Mass Scale (Industry Perspective)
From police chiefs and unions to civil rights advocates and President Obama, a national consensus is emerging that we need to deploy police body-worn cameras as fast as possible to make policing safer and fairer for all.
But considering that there are 800,000 state, local and federal law enforcement officers who might wear such cameras, deploying them on a vast scale will be very challenging for both technical and policy reasons.
First, only the cloud is large enough to store all the data that video police cameras will generate
The most obvious problem is where to put all the video the devices will produce. A single officer equipped with a camera may generate a terabyte of data per year. Even more data will come from dash cams, interrogation room cameras and fixed surveillance devices.
The only realistic place to store all this video is in the cloud. Many police departments using body cameras today operate storage systems on their own premises. But everyone knows this approach can’t scale. The leading camera vendors like Taser and Vievu already work with cloud providers such as Amazon and Microsoft to bundle “all you can eat” cloud storage services with their devices. Costs for a barebones configuration typically run $50 or more per month per user, including both the device and cloud storage.
Second, police departments will struggle to manage huge video data volumes – automation will be essential
But a much bigger problem than storage looms on the horizon. The purpose of body-worn cameras is not to fill petabytes and exabytes of disk space in football-field-size data centers. The goal is to improve interactions between the police and the public they serve. To justify its cost, law enforcement agencies must be able to filter through footage quickly and effectively. They need to review it for investigative, training and disciplinary purposes. They need to share it with fellow agencies, prosecutors and defense lawyers. Last but not least, they need to be able to disclose it – at least selectively – to the public and the media. All this will have to happen while guaranteeing chains of custody, ensuring that only authorized users have access, and protecting the privacy of citizens and officers.
The fundamental problem that police departments gathering large amounts of video face is that the daily tasks they need to perform with this video are labor-intensive. Searching through thousands of hours of video, transcribing and indexing what is said in them, blurring the faces of citizens or officers to protect their privacy – these tasks are impossible to perform at scale without assistance from powerful automation tools.
The only way to manage such immense quantities of data is with machine learning, an advanced form of software that can perform tasks that until now only humans could do. Examples include understanding speech and recognizing human faces or other complex shapes. Researchers are already working to bring machine learning tools to law enforcement agencies to help manage their body-worn camera footage. While many firms are working in this area, here we consider two examples under development.
The first example of what machine learning can do is the automated transcription and indexing of the spoken words captured in police videos. Departments that accumulate thousands of hours of video archives need practical ways of searching this video, or it will be of little use. Linking a time-indexed transcript of spoken words with the video stream makes search fast and accurate. In the next few years, scene analysis algorithms will go a step further and automatically generate simple textual descriptions of the objects, people and events recorded by videos.
A second example is the automated redaction of sensitive information in images prior to their public release. All agree that police videos should not be broadcast to the world without safeguards. But what will the rules be and who will make them? The ACLU and other advocates rightly insist that the privacy of citizens captured on video must be protected. Police unions have also raised valid questions of their own about officers’ privacy.
Machine learning offers a powerful and radical alternative to labor-intensive video redaction tools. Users simply tell the software what they want to redact – human faces, badge numbers, license plates – and let the technology do the work automatically. The savings in time and labor resulting from this approach are dramatic.
Smart software by itself cannot solve complex social problems. But it can help ensure that the coming mass deployment of body-worn cameras makes policing more effective and safer for all concerned. As elected officials, police departments and communities across the country grapple with the policy challenges of police video, they should take the time to educate themselves about the new possibilities that innovations like machine learning bring.
Source: Making Police Body Cams Work on a Mass Scale (Industry Perspective)
Via: Google Alert for ML