Smart Events Detector Plugin

One Step Further Towards Real-Time Driving Maneuver Recognition Using Phone Sensors

About

This plugin deals with the problem of driving maneuvers detection, based on smartphones sensors, especially the accelerometer, the gyroscope and GPS sensors. A framework based on a number of deep learning methods, for maneuvers classification and clustering, is introduced. Maneuvers were classified into 15 major classes like:

• Idle
• Lane Change Left
• Lane Change Right
• Obstacle Avoid Left
• Overtake Left
• 45-Turn Left
• 45-Turn Right
• 90-Turn Left
• 90-Turn Right
• 180-Turn Left
• Noise

We selected three classifiers, each offering good performance for recognizing our set of activities, and investigated how to combine them into an optimal set of classifiers.

Dataset

Use our dataset

We gathered the original data: it is about 1000 kilometers of video and telemetry data gathered with a phone on the roads.
You can freely use it to get start with our plugin.

Use your own dataset

In this case you should upload a csv file describe the files distribution in your dataset directory.
- ie -  Type, Date, Time, Extension, Absolute path

Where Type can take 3 values:

• Sensors

  • You need to have a csv data file in the format specified in the "Csv file format" section (Overview).

• Events

  • If you need to generate training set after, then you should create training set with the help of dashboard. It means you can record events and save it in json format for further training of model.
  • But it's not necessary if you just want predict this data.

• Video

  • Video recorded at the same time of csv file.
  • Subtitles file (srt) for the video. It will only display the data for the period of video.

Example:

The following image represents files management of the day "2015-01-25".

Events,2015-01-25,16:29:16,json,/Users/Ramah/Data/Events/InputFolder/20150125/20150125162916_60.07427_30.34051_1000057.json
Events,2015-01-25,16:32:38,json,/Users/Ramah/Data/Events/InputFolder/20150125/20150125163238_60.07657_30.33811_1000057.json
Events,2015-01-25,16:36:00,json,/Users/Ramah/Data/Events/InputFolder/20150125/20150125163600_60.07689_30.33703_1000057.json
Events,2015-01-25,16:39:23,json,/Users/Ramah/Data/Events/InputFolder/20150125/20150125163923_60.07679_30.33816_1000057.json
Events,2015-01-25,16:42:45,json,/Users/Ramah/Data/Events/InputFolder/20150125/20150125164245_60.07564_30.34149_1000057.json
Events,2015-01-25,00:00:00,json,/Users/Ramah/Data/Events/InputFolder/20150125/all.json
Sensors,2015-01-25,00:00:00,csv,/Users/Ramah/Data/Sensors/2015-01-25_SensorDatafile.csv
Video,2015-01-25,16:29:16,mp4,/Users/Ramah/Data/Video/20150125/20150125162916_60.07427_30.34051_1000057.mp4
Video,2015-01-25,16:29:16,srt,/Users/Ramah/Data/Video/20150125/20150125162916_60.07427_30.34051_1000057.srt
Video,2015-01-25,16:32:38,mp4,/Users/Ramah/Data/Video/20150125/20150125163238_60.07657_30.33811_1000057.mp4
Video,2015-01-25,16:32:38,srt,/Users/Ramah/Data/Video/20150125/20150125163238_60.07657_30.33811_1000057.srt
Video,2015-01-25,16:36:00,mp4,/Users/Ramah/Data/Video/20150125/20150125163600_60.07689_30.33703_1000057.mp4
Video,2015-01-25,16:36:00,srt,/Users/Ramah/Data/Video/20150125/20150125163600_60.07689_30.33703_1000057.srt
Video,2015-01-25,16:39:23,mp4,/Users/Ramah/Data/Video/20150125/20150125163923_60.07679_30.33816_1000057.mp4
Video,2015-01-25,16:39:23,srt,/Users/Ramah/Data/Video/20150125/20150125163923_60.07679_30.33816_1000057.srt
Video,2015-01-25,16:42:45,mp4,/Users/Ramah/Data/Video/20150125/20150125164245_60.07564_30.34149_1000057.mp4
Video,2015-01-25,16:42:45,srt,/Users/Ramah/Data/Video/20150125/20150125164245_60.07564_30.34149_1000057.srt

Prepare your dataset for training and validation

At this stat we normalize your raw data & generate csv files from it and hand made event files (json).

• OutTrain.csv from selected days as training set, thas you can download it, and start your model.
• OutCv.csv from selected days as validation set, thas you can download it, and validate your trained model.

Note : we get the raw data from the Sensors directory and json files in the Events Directory

Generate Testset

generate csv files from raw data.

Predict events

Predict events from generated csv files (Testing set) in run page.