I need an algorithm, which will count bicycles and people crossing regular camera's view. In other words, camera will be facing bicycle lane and some part of pedestrian sidewalk. The algorithm must run on Jetson tx2 device. The output of the algorithm should be an API, which will return the number of detected bicycles and pedestrians in close to real time. I believe it is the best to use Tensorflow for the task, however, if you believe that some other frameworks may do the job better, please go ahead for this.
In other words, tasks are like these:
Bicycle and pedestrian detection and counting. Signal will come via LAN cable from camera to Jetson TX2. With possibility to adjust counting/detection zone. Accuracy should be around 95% for bicycles. For people could be around 90%.
• Results are returned with timepstamp. Ideally, results should be written into some tiny database, which could be later easily accessed by us to export or read.
• Code should be compatible with Jetson TX2 architecture running on Nvidia embedded into Jetson.
• You provide us with the source code and we compile it to Jetson ourselves.
Less than 30 hrs/week
Less than 1 month< 1 monthProject LengthDuration
I am looking for a mix of experience and value