Real-Time Transport Tracking in Mobile Apps

Мы решаем проблему точного трекинга транспорта в реальном времени

We solve the problem of precise real-time transport tracking.

Our engineers have faced situations where a GPS tracker on a bus showed a route jumping along parallel streets, and the dispatcher could not figure out where the vehicle actually was. The driver had already passed the stop, but on the map it was still approaching. This is not a hardware issue but a fault in filtering and map matching. We develop tracking systems that use data from hardware trackers (Teltonika, Queclink, Galileosky) or GTFS Realtime and overlay them on the map considering speed, route, and terrain. Experience with fleets from 50 to 1000 vehicles allows us to guarantee positioning accuracy with an error of no more than 5 meters.

What data sources do we use?

For logistics companies and fleet management, the transport device is a hardware GPS tracker (Teltonika FMB920, Queclink GV350, Galileosky), not the driver's phone. The tracker sends packets via MQTT or HTTP protocol to the server. The app serves only as a data visualizer, not as a source.

If the task is to display public transport using GTFS Realtime (Google standard), the source is the open APIs of city transport operators. The format is Protocol Buffers (transit_realtime.FeedMessage), parsing via the gtfs-realtime-bindings library.

For apps using the driver's phone (taxi, corporate transport) — the same stack as courier tracking but adjusted for movement speed.

Comparison of tracker types

GPS filtering at high speeds: how to eliminate drift?

At speeds of 80-120 km/h, horizontal GPS drift is less than in the city with high-rises, but another problem: during sharp turns, the marker may 'overtake' the actual vehicle position due to update delay. A Kalman filter smooths out this delay.

Simple Kalman filter implementation for coordinates in Kotlin:

class KalmanFilter(private var accuracy: Float = 1f) { private var lat = 0.0 private var lon = 0.0 private var variance = -1f fun process(lat: Double, lon: Double, accuracy: Float, timestamp: Long): LatLng { if (variance < 0) { this.lat = lat; this.lon = lon; variance = accuracy * accuracy } else { val dt = (timestamp - lastTimestamp) / 1000f variance += dt * 3f * 3f // rate of change 3 m/s val k = variance / (variance + accuracy * accuracy) this.lat += k * (lat - this.lat) this.lon += k * (lon - this.lon) variance *= (1 - k) } lastTimestamp = timestamp return LatLng(this.lat, this.lon) } private var lastTimestamp = 0L } 

On iOS, a similar implementation in Swift or using CLLocationManager with CLActivityType.automotiveNavigation — Apple applies its own filter.

Map matching: which service to choose?

A scheduled bus travels along a fixed route — GPS points between stops must lie strictly on that route, not jump to a parallel street. Map matching: we take a sequence of GPS points and 'snap' them to the nearest road graph segment.

OSRM self-hosted: GET /match/v1/driving/{coordinates}?radiuses={radiuses}&geometries=geojson. Returns a matched track with waypoints. According to OSRM documentation, latency when self-hosting is < 20 ms, which is acceptable for real-time.

Google Roads API snapToRoads — easier to integrate but paid ($5 per 1000 calls) and limited to 100 points per request.

Comparison OSRM vs Google Roads API

Conclusion: for frequent use (1000+ vehicles), OSRM self-hosted is 250 times cheaper and 5 times faster.

How to ensure smooth movement animation?

The bus/truck marker is a custom PNG or SVG with rotation according to the direction of movement. Direction in degrees: atan2(dLon, dLat) * 180 / PI. On Android — BitmapDescriptorFactory.fromBitmap(rotatedBitmap) with rotation via Matrix.postRotate(). On iOS — GMSMarker with iconView, rotation via CGAffineTransform(rotationAngle:).

The route is a polyline. Google Directions API for calculation or pre-saved GTFS shapes.txt. On the map — GMSPolyline / Polyline with custom color and width. The traveled section is a different color (e.g., gray instead of blue).

Movement animation — interpolation between points. The GPS tracker update interval is usually 30-60 seconds, not 5. This means the marker should smoothly move for 30 seconds from one point to the next, not jump. ValueAnimator on Android with LinearInterpolator, CADisplayLink on iOS.

Server side: how to ensure scalability?

For a fleet of 50-200 vehicles — Socket.IO or WebSocket on Node.js. The server stores current positions in Redis with TTL. The client subscribes to the fleet/updates channel and receives updates for all vehicles in a batch every 10-15 seconds instead of individual events — saving 80% traffic.

For large fleets (1000+ vehicles) — MQTT broker with topics vehicle/{id}/gps. The client subscribes only to vehicles of interest.

Historical route storage: PostgreSQL + PostGIS for geospatial queries (“show all vehicles that passed through zone X yesterday”), TimescaleDB for time-series metrics (speed, fuel).

What is included in the work?

• Data source audit: tracker/phone/GTFS — determine type and protocol.
• Server bus design and stack selection (MQTT/WebSocket/Redis).
• Mobile client implementation: markers, routes, animation, states.
• Map matching integration (OSRM self-hosted or Google Roads).
• Load testing with traffic simulation up to 1000 vehicles.
• API and configuration documentation.
• Dispatcher training (1 hour).
• 2 weeks of support after launch.

Development timeline — from 2 to 6 weeks depending on data source, number of platforms, and fleet size. Certified engineers with 5+ years of experience guarantee stable operation. Average implementation cost for a fleet of 20 vehicles starts from 500,000 rubles. We'll evaluate your project for free — contact us! Request tracking development — contact us for a consultation.