Release notes

We are pleased to announce the availability of version 1.11.1 Live Sense SDK for iOS.

This is a stable release. We are constantly working on new features, stabilizing APIs, improving our KPIs, and evaluating your feedback.

Changes

This section documents major changes to each release that may require you to update your application.

Version 1.11.1

Improvements

• Improved the detection of stop signs by Road Signs model.

Known issues and workarounds

• The distance and relative position calculation for detections is in Beta, the returned distance and relative position values may be inaccurate.
• Using a lower confidence threshold will result in more detections, but an overall lower accuracy. This can sometimes cause the labels to show false positives. For this reason, applications should keep their confidence threshold at or above 0.6 to maintain reasonably accurate output.
• Lower accuracy for detections is noted for quant variants of pothole and cone_barrier models.
• The Observation library has limited functionality due to the HERE Data Hub shutdown.

Changes from previous releases

This section documents major changes from past releases. We encourage you to use the latest release, 1.11.1.

Version 1.11.0

New features

• New Road Signs model with additional sign types:
  • Advisory speed limits
  • No access
  • Pedestrian crossing
  • Parking restriction
  • Stop here on red
  • No left turn
  • No U turn
  • No U or left turn
  • No right or left turn
• Added support for dynamic region of interest derived from lane markings
• Added lane departure alerts

Improvements

• Improved the the distance and relative position calculation for detections (Beta).
• Updated speed bumps quant model

API changes

• Added - Added determineLaneAlerts and onLaneAlerts to LSDAlertManager
• Added - Added LANE_DEPARTURE to LSDAlertType
• Added - Added bool useLaneDeparture to LSDAlertSettings (default false)
• Added - Added Roiand RoiType
• Added - Added setRoi to LSDAlertManager

Functional and behavioral changes

• Updated entry alerts from LSDAlertManager with support for the following detection classes
  • "car", "rider", "bus", "truck", "bicycle", "motorcycle"

Known issues and workarounds

• The distance and relative position calculation for detections is in Beta, the returned distance and relative position values may be inaccurate.
• Using a lower confidence threshold will result in more detections, but an overall lower accuracy. This can sometimes cause the labels to show false positives. For this reason, applications should keep their confidence threshold at or above 0.6 to maintain a reasonably accurate output.
• Lower accuracy for detections is noted for quant variants of pothole and cone_barrier models.

Changes from previous releases

This section documents major changes from past releases. We encourage you to use the latest release, 1.11.0.

Version 1.10.0

New features

• Added relative velocity calculation for Roadbasics model classes(except traffic-light).

Improvements

• Improved TTC alert.

API changes

• Added RelativeVelocityEstimator
• Added relativeVelocity property to LSDObjectRecognition
• Added collisionBufferDistance property to LSDAlertManager

Known issues and workarounds

• The distance and relative position calculation for detections is in Beta, the returned distance and relative position values may be inaccurate.
• Using a lower confidence threshold will result in more detections, but an overall lower accuracy. This can sometimes cause the labels to show false positives. For this reason, applications should keep their confidence threshold at or above 0.6 to maintain reasonably accurate output.
• Lower accuracy for detections is noted for quant variants of pothole and cone_barrier models.

Changes from previous releases

Version 1.9.0

New features

• Added distance and relative position calculation (Beta) for detections from the Road Signs, Road Hazards, and Brake Light models.
• Added default ML Engine selection mechanism
  • Added setUseDefaultOptions(val:Bool) of ModelOptions

Improvements

• Improved the Road Basic model for TensorFlow Lite.
• Improved the Lane marking model for TensorFlow Lite.
• Improved TTC alert.

API changes

• Added distance and relative position calculation (Beta) for detections from the Road Signs, Road Hazards, and Brake Light models.

Known issues and workarounds

• The distance and relative position calculation for detections is in Beta, the returned distance and relative position values may be inaccurate.
• Using a lower confidence threshold will result in more detections, but an overall lower accuracy. This can sometimes cause the labels to show false positives. For this reason, applications should keep their confidence threshold at or above 0.6 to maintain reasonably accurate output.
• Lower accuracy for detections is noted for quant variants of pothole and cone_barrier models.

Version 1.8.0

New features

• Migrated the ml-lib to TFLite from CoreML.

API changes

• Removed the following classes:

  • LSDVideoCaptureViewRecognition
  • LSDVideoReaderView
  • LSDRecognition
  • LSDVideoReaderViewStatisticsDelegate
  • LSDVideoCaptureStatisticsDelegate
  • LSDVideoReaderViewRecognitionDelegate

• Added initiateDetectionRequest(isParallel: Bool, forModels [LiveSenseDetection.LivesenseModel], completionHandler: @escaping (Bool) -> Void) for initialising models.

• Added LivesenseModel class as a base class for all models.
  • RoadBasicsModel
  • RoadSignsModel
  • RoadHazardModel
  • LanesModel
  • RTGModel
• Added the following classes for object tracker:
  • Manager for Live Sense trackers.
• Added the sendTrackedResults(result : DetectionResult) method to get tracked recognition results.
• Added the sendRecognitionResults(result : DetectionResult) method to get recognition results.
• Added ModelOptions for common options for Live Sense models.

Version 1.7.0

New features

• Added sessionId field to ObservationData.
• Added Pedestrian TTC detection alert.

Improvements

• Improved RoadSigns model for detecting speed-limits and stop-signs.

API changes

• Removed the RoadSigns regional (Brazil, Oceania and South-East Asia) models.

Version 1.6.0

New features

• Added Lanes model that detects lanes on the highways.
• Added RTG model (Beta) for Real Time Guidance that detects class stop-sign, traffic-light.
• Added delegate method lanePoints(segmenntationResult : [LanePointWrapper]) for lanes output.
• Added estimateDistance(recognitions:frameSize:sensorRotation:focalLength:sensorSize:) to get distance of detected object.
• Added determineAlerts(recognitions:frameSize:sensorRotation:focalLength:sensorSize:) to get alerts for recognitions.
• Added distance and relative position calculation (Beta) for detections from the Road Basics model.

Improvements

• Improved Road Signs model classifier.

API changes

• Updated RoadSigns model labels to GFR format.
• Updated all model labels to be consistent across platforms.
• Updated StaticDistanceEstimator and LSDAlertManager as public APIs

Version 1.5.0

New features

• Improved model for detecting brake-lights

Improvements

• Road Hazard Sign label change : cross-walk

Version 1.4.0

New features

• Added a Road Basic (default) model which introduces a new class rider over the existing classes. rider is defined as a person driving a cycle, bicycle, scooter, or a motorbike.
• Added a Road Basic night model to detect features in low light conditions. Supported features - car,motorcycle,person,bus,truck.
• Update to the end-user consent message.

Improvements

• Introduced improvements in plotting the bounding box for detections

API changes

• Exposed distance in LSDAlert to return distance to collision
• Added videoCaptureView.setSideCropPercent to crop input frame in percentage from all/any of the 4 sides
• Added videoCaptureView.setUseIncrementalCenterCrop to enable/disable incremental zoom focused at the center

Version 1.3.0

New features

• Added an improved Road Signs global model with stop signs included. This contains maximum speed limit signs for all supported regions which include NA, Brazil, SEA, U,K and Oceania.
• Introduced improvements to calculation of Time to Collision (TTC).

API changes

• Added StaticDistanceEstimator - to calculate distance of an object from the camera.
• Added videoCaptureView.locationUpdates() to start and stop location updates with required level of configuration
• Changed AlertType to LSDAlertType and the following enums are now available: {ENTRY,PROXIMITY,PROXIMITY_VEHICLE,CHANGE_DEVIATION,UNDEFINED}
• Changed AlertSeverity to LSDAlertSeverity and the following enums are now available: {INFO,NORMAL,WARNING,DANGER,UNDEFINED}
• Added LSDAlertManager to receive TTC alerts based on severity levels.
• Updated the following methods in LSDVideoCaptureStatisticsDelegate:
  • getVideoStats - Added droppedFrames as in input parameter
  • Added output(sampleBuffer:CMSampleBuffer,videoOutput:AVCaptureVideoDataOutput) method
  • Changed alertReceived return type from object to array of LSDAlert
• Renamed LSDCaptureView to LSDVideoCaptureViewRecognition

Improvements

• Improved the detection accuracy of Live Sense models in portrait mode for iOS 13 and above.

Version 1.2.0

New features

• New detection capabilities available for beta testing:
  • Traffic Light Status - this detects the state of the traffic light: red, orange or green.
  • Height Restriction Signs - this detects road signs that have height specified in feet and inches.
  • Tunnels and Overpasses - this detects tunnels and overpasses that can set a height restriction for permitted vehicles. Can be used to calculate the permitted height limit.

API changes

• Added func updateClassConfidence(classLabel : ClassLabelName, classConfidence : Float) - Configure minimum detection confidences separately for each individual class in AI/ML models that will supersede confidence configured for parent model.

Resolved issues

• Set confidence for pedestrian class to 0.45 to improve detection of pedestrians which are closer to the device camera.

Improvements

• Improved the detection accuracy of Road Signs model.

Version 1.1.0

New features

• Observation Library - Features detected by Live Sense SDK can now be saved to HERE Data Hub. These features can then be retrieved from Data Hub via the SDK.
• New detection capabilities:
  • Speed bumps - White strips

API changes

• The following methods were added to initialize, save and retrieve features from HERE Data Hub:

  • LSOClient.init(spaceID:String,auth:String){}
  • LSOClient.validateCredentials(spaceId: String, authToken:String,output: @escaping (LSOResponseHandler)){}
  • LSOClient.putObservations(_ observation: LSOData, success: @escaping (LSOSuccessHandler), failure: @escaping (LSOFailureHandler)){}
  • LSOClient.getObservations(_ position: CLLocation,to:Int,from:Int,success: @escaping (LSOSuccessHandler),failure: @escaping (LSOFailureHandler)){}