Utilizing Amazon Rekognition to enhance bicycle security

Biking is a enjoyable strategy to keep match, take pleasure in nature, and join with associates and acquaintances. Nonetheless, driving is turning into more and more harmful, particularly in conditions the place cyclists and automobiles share the highway. In response to the NHTSA, in the USA a median of 883 individuals on bicycles are killed in site visitors crashes, with a median of about 45,000 injury-only crashes reported yearly. Whereas complete bicycle fatalities solely account for simply over 2% of all site visitors fatalities in the USA, as a bicycle owner, it’s nonetheless terrifying to be pushed off the highway by a big SUV or truck. To higher shield themselves, many cyclists are beginning to trip with cameras mounted to the entrance or again of their bicycle. On this weblog submit, I’ll show a machine studying resolution that cyclists can use to raised determine shut calls.

Many US states and nations all through the world have some form of 3-feet regulation. A 3-feet regulation requires motor autos to offer about 3 toes (1 meter) of distance when passing a bicycle. To advertise security on the highway, cyclists are more and more recording their rides, and in the event that they encounter a harmful scenario the place they aren’t given an acceptable secure distance, they will present a video of the encounter to native regulation enforcement to assist right habits. Nonetheless, discovering a single encounter in a recording of a multi-hour trip is time consuming and sometimes requires specialised video abilities to generate a brief clip of the encounter.

To resolve a few of these issues, I’ve developed a easy resolution utilizing Amazon Rekognition video evaluation. Amazon Rekognition can detect labels (primarily objects) and the timestamp of when that object is detected in a video. Amazon Rekognition can be utilized to shortly discover any autos that seem within the video of a recorded trip.

If a bicycle owner’s digicam data a passing car, it should then decide if the car is simply too near the bicycle—in different phrases, if the car is inside the 3-foot vary set by regulation. Whether it is, then I need to generate a clip of the encounter, which could be supplied to the related authorities. The next determine exhibits the view from a bicycle owner’s digicam with bounding bins that determine a car that’s passing too near the bicycle. A field on the backside of the picture exhibits the approximate 3-foot space across the bicycle.

Answer overview

The structure of the answer is proven within the following determine.

The steps of the answer are:

1. When a bicycle owner completes a trip, they add their MP4 movies from the trip into an Amazon Easy Storage Service (Amazon S3)
2. The bucket has been configured with an S3 occasion notification that sends object created notifications to an AWS Lambda
3. The Lambda operate kicks off an AWS Step Capabilities workflow that begins by calling the StartLabelDetection API as a part of Amazon Rekognition movies. The StartLabelDetection API is configured to detect Bus, Automobile, Fireplace Truck, Pickup Truck, Truck, Limo, and Transferring Van as labels. It ignores different associated non-vehicle labels like License Plate, Wheel, Tire, and Automobile Mirror.
4. The Amazon Rekognition API returns a set of JSON figuring out the chosen labels and timestamps of detected objects.
5. This JSON result’s despatched to a Lambda operate to carry out the geometry math to find out if a car field overlapped with the bicycle secure space.
6. Any detected encounters are generated and handed off to AWS Elemental MediaConvert, which may create snippets of video comparable to the detected encounters, utilizing the CreateJob API
7. MediaConvert creates these movies and uploads them to an S3 bucket.
8. One other Lambda operate known as to generate pre-signed URLs of the movies. This permits the movies to be briefly downloaded by anybody with the pre-signed URL.
9. Amazon Easy Notification Service (Amazon SNS) sends an e-mail message with hyperlinks to the pre-signed URLs.

Stipulations

To make use of the answer outlined on this submit, it’s essential to have:

1. An AWS account with acceptable permissions to let you deploy AWS CloudFormation stacks
2. A video recording in MP4 format with the .MP4 extension utilizing the H.264 codec. The video must be from a entrance or rear-facing digicam, from any off-the-shelf vendor (for instance GoPro, DJI, or Cycliq). The utmost file dimension is 10 GB.

Deploying the answer

1. Deploy this resolution in your surroundings or choose Launch Stack. This resolution will deploy within the AWS US East (N. Virginia) us-east-1 AWS Area.

2. The Create stack web page from the CloudFormation dashboard seems. On the backside of the web page, select Subsequent.
3. On the Specify stack particulars web page, enter the e-mail handle the place you’d wish to obtain notifications. Select Subsequent.
4. Choose the field that claims I acknowledge that AWS CloudFormation would possibly create IAM assets and Select Subsequent. Select Submit and the set up will start. The answer takes about 5 minutes to be put in.
5. You’ll obtain an e-mail confirming your Amazon SNS subscription. You’ll not obtain emails from the answer except you affirm your subscription.
6. After the stack completes, choose the Outputs tab and pay attention to the bucket title listed beneath InputBucket.

Utilizing the answer

To check the answer, I’ve a pattern video the place I requested a stunt driver to drive very carefully to me.

To start the video processing, I add the video to the S3 bucket (the InputBucket from the Outputs tab). The bucket has encryption enabled, so beneath Properties, I select Specify an encryption key and choose Use bucket settings for default encryption. Selecting Add begins the add course of, as proven within the following determine.

After a second, the step operate begins processing. After a couple of minutes, you’ll obtain an e-mail with hyperlinks to any encounters recognized, as proven within the following determine.

In my case, it recognized two encounters. Within the first encounter recognized, I rode too near a parked automotive. Nonetheless, within the second encounter recognized, it exhibits a harmful encounter that I skilled with my stunt driver.

Had this been an precise harmful encounter, the video clip may very well be supplied to the suitable authorities to assist change habits and make the highway safer for everybody.

Pricing

As a result of it is a absolutely serverless resolution, you solely pay for what you employ. With Amazon Rekognition, you pay for the minutes of video which are processed. With MediaConvert, you pay for normalized minutes of video processed, which is every minute of video output with multipliers that apply primarily based on options used. The answer’s use of Lambda, Step Capabilities, and SNS are minimal and can possible fall beneath the free tier for many customers.

Clear up

To delete the assets created as a part of this resolution, go to the CloudFormation console, choose the stack that was deployed, and select Delete.

Conclusion

On this instance I demonstrated the way to use Amazon Rekognition video evaluation in a novel situation. Amazon Rekognition is a strong pc imaginative and prescient software that means that you can get insights out of pictures or video with out the overhead of constructing or managing a machine studying mannequin. After all, Amazon Rekognition may deal with extra superior use circumstances than the one I demonstrated right here.

On this instance I demonstrated how utilizing Amazon Rekognition with different serverless companies can yield a serverless video processing workflow that—on this case—might help enhance the protection of cyclists. Whilst you may not be an avid bicycle owner, the answer demonstrated right here could be prolonged to quite a lot of use circumstances and industries. For instance, this resolution may very well be prolonged to detect wildlife on nature cameras or you may use Amazon Rekognition streaming video occasions to detect individuals and packages in safety video.

Get began at present by utilizing Amazon Rekognition on your pc imaginative and prescient use case.

Concerning the Writer

Mike George is a Principal Options Architect at Amazon Internet Providers (AWS) primarily based in Salt Lake Metropolis, Utah. He enjoys serving to prospects resolve their know-how issues. His pursuits embrace software program engineering, safety, synthetic intelligence (AI), and machine studying (ML).