Software Changelog

Version:

v2.8.0

Date:

TBD

Added

• PCAP Recorder module to the Event Server that triggers PCAP recordings based on Event Server Modules

• Setting for pcap_start_buffer_duration_s to control the duration of time added to the beginning of PCAP recordings when buffer is enabled

• Option to disable publishing object points

• Ability to disable background filter affect for people and vehicle classifications separately

• Enhanced telemetry endpoint with new pcap_recordings object containing detailed recording state information including recording state, start timestamp, duration, and filename for all active recordings

• Variable object drop time calculation to improve identity switches

• New default data association thresholds based on classification

• Numeric control for DL AOI ground height on AOD setup tab

• Sensor info label on hover when using selection tool in setup tab

• Hold ‘S’ key to toggle sensor selection tool with info label toggle support

• Switch to selected sensor’s image if available in setup tab

• Preview source in Viewer mode when hovering over 2D viewer

• Ability to resize left and right panes in Detect Viewer

• Expose additional information during sensor select: port numbers, hostname, IP, MAC Address, Sensor Model Name and Part Number available on sensor hover in Setup tab

• Expose ‘User-Defined-Classification’ Tracked Object Label

• Expose ‘total_chained_object_length’ Traced Object Label

• Show segment length when drawing zones

• Ability to resize near IR image via dragging bottom bar

• Add zone occupancy count visibility icon and default to false

• All background clouds appear at once when loading Detect Viewer

• Add search bar for processed cloud list, source list, zone, and diagnostics tab

• Add field convert_to_geo_coordinates to the default publishers settings

Changed

• Changed timestamps on diagnostics page to reflect uptime instead of last message time

• buffer_length_s query parameter for /perception/api/v1/recording/start_buffer endpoint is now optional

• Event zone occupancy multithreaded to improve performance with large number of zones

• Cluster downsampling now downsamples to exactly the requested max points

• On startup the system will keep trying to connect to sensors indefinitely

• Enable ahead-id, behind-id, and sub-classification labels by default

• Persist zone label visibility, defaults to not visible

• Persist selected tab across page refreshes

• Hide transform axis-translate gizmos when angle with associated axis is too small

• Only show DL AOD of enabled multi sensor inference AODs unless they are selected

• Cylinder scale from dynamic to fixed for Person classification

• Visibility for Object UUID on highlight to false on default and move tracked object labels, sensor info label out of developer mode. Only show 5 digit of the UUID, click to expand and copy full UUID

• Move cylinder bounding box control to leftmost and restricted its availability to Person classification only

• Unsuccessful addition of sensors, have been promoted to Critical Alert status

• More vibrant Error Color, increased unsuccessful messages to 10secs, uppercase ERROR word in feedback messages

• Changed the debian and RPM packages to not attempt to pull docker images during installation or upgrade if the images already exist on the system

• To associate ahead and behind IDs for trucks and trailers, a gap between the truck and trailer is now allowed if both vehicles are moving above a certain velocity

• Separated object NMS settings from cluster NMS settings

• Modified object NMS stage to not allow people objects to suppress vehicle objects

• Add DL AOD’s to event recordings

Fixed

• Reduced the logging spam from when a receive error occurs when listening for a message on a channel in Event Server

• Invalid classifications were allowed in the configuration which defaulted to unknown objects. This has been fixed and invalid classifications will now cause a configuration error in Event Server.

• Cluster merge would result in incorrect number of foreground points very rarely resulting in cluster filtering not respecting setting

• Fixed issue where sensors could not be added or removed during a PCAP recording

• Fixed occupation server blocking updating statistic on one sensor at a time which could result in slow update_occupations

• Fixed any settings changed was causing all DL models to re-init even if it was unrelated

• Fixed issue where saving of the background at the end of a PCAP recording would result in dropped data

• Background points of range 0 were sometimes getting sent even though they are not useful

• Clear failed lidar connection alert whenever we connect to a lidar

• Distance constraint toggle on DL tab only applied after tab changed

• Fixed incorrect “invalid zone” error that occasionally appears during zone creation

• Clean up interactions with Network Tab and fixed an issue where routing priority would not be set

• Hide ‘View’ and ‘Download Logs’ buttons in the Cloud Viewer since they require direct edge access

• Remove double copyright in header

• Fix Cloud Viewer auth error when stopping event recordings

• Cloud Viewer keep alive timer not decreasing

• Fix GUI to not show recording status for background recordings

• Fix frontend stop wrong recording when there are multiple recordings

• Fix recording timer resetting when refreshing the browser

• Fix point cloud height exclusion/filtering in Viewer tab

• Fix cannot collapse the subsection in diagnostic tab

• Not requesting the UDID and Serial Number when not on BlueCity

• Change style of the measurement label and show measure unit based on preferences

• Fix trace of old object issues

• Fix a bug that caused the pasted zone to briefly disappear and reappear. This change improves the stability of the copy-and-paste functionality

• Fixed the uninstallation and purge process for the Ouster package. Previously, a cache directory located at /opt/ouster would remain after uninstalling the package, requiring manual removal. Now, this directory is automatically removed, ensuring a cleaner and more complete uninstallation

• Fixed hostname resolution endpoint to prefer non link-local addresses when multiple addresses are available

• VisitAggregator was using the object’s last_update timestamp instead of the frame_timestamp from the occupations. This meant that objects that have not elapsed the debounce time were still considered for events

• The point counts were sometimes zero for a zone even though there were points in the zone. This was due to a race condition. Now if the processing is not finished, it will use the previous point counts

• Object Ahead and Behind IDs could associate with a non-vehicle object in specific cases

• Object NMS now clears the ahead and behind ids if the associated object is removed

• Ahead and behind IDs could be de-synchronized between the two associated objects in some cases

• Ahead and behind IDs could be swapped in some cases, resulting in the front object being labeled as a trailer and the rear object being labeled as a truck

• Ahead object could be an UNKNOWN classification if its subclassification was truck

• Block the use of link-local IP addresses for sensors as they are not supported

• Event Recorder Timestamp Buffer performance improvements in attempts to eliminate empty recordings and recordings missing data

• End Timestamp is not set when a User Event Recording is stopped without a Stop Trail Seconds

• Get MAC address for Interface no longer case-sensitive

• Fixed an issue where the installation scripts would hang if the user had 2.16.0 Docker Compose or 23.0.0 Docker installed

• Disable start update button during Unpacking of software update

• File validation for software update selection

• East/west inversion in compass component

Removed

• Removed multiple Event Recorder settings options typically not used

• Removed Front/Back AOD option

• Removed constraint for ML AOD’s to be within 20m of sensor

Known Issues

• Enrichments can appear out of sync with live data

• Routing priority on Network tab is not reported properly when interface is disconnected

• Velocity of stationary vehicles may show up to 4m/s

• Walls detected as Large Vehicles

• Nodes cannot be the first source added to a system in distributed perception - a sensor must be added first

• Nodes added at the same time as a sensor will not show up in the sources list on the Setup tab. Refreshing the browser tab is required to show the sensor and node

• Agent API can be slow to respond requiring at least 1 second timeout

• Setting image_publish_period to 0 prevents enabling image publishing again. Work around is to restart the perception container

• Discovered sensor list never removes unplugged/un-powered sensors

• Adding sensors to a node and aggregator at any point during a distributed perception setup can cause unexpected behavior due to conflicting extrinsic parameters

Version:

v2.7.1

Date:

2025-06-24

Fixed

• Issue where specific orientations of objects could result in an NaN value for those bounding boxes

Version:

v2.7.0

Date:

2025-06-17

Added

• GoogleMaps integration allowing users to choose lat/long coordinates and perform alignment directly on map underlay

• Gemini Events triggering for the following situations:

  • Health monitoring for the system reporting lidar, software and system metrics

  • Intrusion module flagging when objects traverse from one zone to another

  • Location presence which can monitor and flag events based on object counts as well as report metrics about queue lengths

  • Visit fanout module which can distinctly identify when objects enters a zone, when an object is in a zone, and when an object exits a zone

  • Speed filter for identifying objects travelling below or above user-defined speed thresholds

  • Object proximity monitor identifying when objects of one class are nearby objects of another

  • Location occupied check which can report presence based on point count zones

  • Object queue monitor colorizing objects in event recordings based on zone presence and whether they’re stationary as well as emphasize certain zones in the scene for context based on colorization

  • MOXA sender supporting OR’ing events from multiple upstream modules to single moxa channels

• Alert when the DL model fails to initialize and when DL inference fails

• Tractor trailers now have ahead and behind ids to associate the parts

• Downsample points sent to inference for an 12% average speedup in DL modes

• Timed_out state in the get_sensor REST call for sensors that have timed out

• Ability to run different ML models with different sensors. Useful for deployments with a mixture of indoor and outdoor areas

• DELETE /network/lan/ endpoint to remove a network interface’s custom configuration (will return to factory defaults)

• Add button to reset the camera position

• Ability to download all logs with 1-click

• Support for easily swapping sensors between lidar ports and auto-connecting to added sensors

• MQTTS support for lidarhub

• Support for live 2D view in Portal through lidarhub

Changed

• Change event recordings to use 8-bit serialization for reduce upload bandwidth and playback load-time

• Stop sending reflectivity data with object and cluster clouds to save bandwidth since they weren’t being used. Bandwidth reduced by 50%

• Filter number of occupations shown based on object visibility for the classification

• Hide transmission delay under a toggle in Preferences

• Service reachability debounce and reset times

• Fixed missing password policy enforcement in reset_user_password script

• /host_address endpoint will now check the mDNS records if the system resolver fails to resolve the hostname

• Reduced false positives with tilt warning alert by increasing the angle threshold for the tilt warning to 10 degrees and default debounce time to 30 seconds

• Changed default zoom level when loading a new scene. GUI will cache the zoom/angle/position of the camera for the next time the user loads the viewer

Fixed

• Clear sensor destination changed alerts if we receive data because the destination must now be correct

• Distributed perception background clouds are now deserialized and published correctly

• Fixed issue with background clouds so they’re always using the latest transform

• No longer send stale images or backgrounds for sensors that have timed out

• Window sometimes becomes scrollable when the window height is too small

• Multiple issues with the world tool

• Sensor IP resolution issues on power-up

• Memory growth issues found in ouster-solutions-agent and perception

• More robust logic for resolving sensor IP addresses on power-up

Version:

v2.5.0

Date:

2025-02-07

Changed

• Improved tracking of fast-moving, turning objects

• Improved truck/trailer turn tracking

• Improved reflectivity image scaling in Detect Viewer

• Improved tracking in z-direction with model update

• Computation Optimization for large number of sensors and clusters

• Ability to select “front-only” for ML AOD

• Changed transmission delay units to seconds

• Navigation between images with left/right arrow keys

• Script to reset RBAC password

• Reduced downtime on software updates to less than a minute (must be running 2.5+ already)

• Computation optimizations

• Fine-grained, per service access control roles. Reduced to view-detect and manage-detect applying to the entire Edge Detect solution.

• The Event zone server container was removed and merged into the ouster perception container. The API paths were left unchanged.

• [LidarHub] Add Network Interface as an Option for Setting the LidarHub Hardware ID

• [LidarHub] Add Enrichments REST API

• [LidarHub] Add Web Socket relay server w/default Enrichments source

• [LidarHub] Add Web Socket Echo web application

• [LidarHub] Changed aggregation_publishing_enabled default value to true on new and existing installs

Fixed

• Near IR channel values on displayed point cloud were incorrect.

• Fixed precision of publishing at desired interval rate

• Fixed objects showing up as background clouds

• Fix crash on auto-highlight feature

• Fixed issue where powered off sensor could not be removed

• Fixed issue where feedback area was clipped in Detect Viewer

• Fixed issue to properly colorize the point cloud in the Detect Viewer by Near-IR

• Fixed issue where sensors were not being listed during event playback

• Fixed 2d Image view not showing in playback mode

• Fixed issue with offline sensors continuing to show in the Source Mangement pane. This list is now refreshed every 5 minutes to remove any sensors which are no longer discoverable.

• [LidarHub] Report removed/deleted Sensors and Event Zones to Portal for immediate update

• [LidarHub] Prevent invalid world settings via the REST API interface and the /map web application

• [LidarHub] Align Event Recording play bar banner with event start and end timestamps

• [LidarHub] Properly indicate Event Recording failure to Portal when retrieval of background point clouds fails

Removed

• object_filters in event zone metadata

Version:

v2.4.1

Date:

2024-11-12

Changes

• New ML model trained on a wider variety of datasets which runs approximately 16% faster than the 2.4.0 model

Fixed

• Fixed dropped messages on heavily loaded GPU systems causing GUI object flickering and reduced accuracy.

Version:

v2.4.0

Date:

2024-10-29

Added

• Multi sensor inference option, default to running on parking settings

• Option to select lidar return to use for all sensors (in perception settings). Defaults to farthest in outdoor profiles, nearest in indoor profiles.

• Zone statistics (min, max, median, upper percentile, and lower percentile) for X, Y and Z axes

• “Download Benchmarks” button.

• Ability to configure DNS through Network tab

• Added Left/Right arrow buttons to cycle through 2D images

• [LidarHub] Settings RPC Integration with Analytic Server

• [LidarHub] Telemetry Integration with Analytic Server

• [LidarHub] Add parameter stop_trail_second to REST API /user_recording/stop

• [LidarHub] Pass user-defined Event Recording Event ID to Portal for display

• [LidarHub] Real-Time and Time Series Aggregation of sub-classifications by zone

• [LidarHub] Trigger Event Recordings on object sub-classification

• [LidarHub] Record Event Recordings in the Past (anytime within the recording buffer)

Changed

• Faster at removing incorrect objects. Objects that are no longer detected and overlap with other objects are removed immediately

• Include_classifications now applies to limit what comes out of the occupations stream

• Default to not allow any overlap between bike detections on fast ml profiles

• Setting max_number_settings no longer required and is ignored if set from old installs

Fixed

• If occupations and tracked object get out of sync due to slow gui the occupations count will now stay correct

• Refind module will no longer match pedestrian/bikes to vehicles in ML modes

• Zero-Length recording now longer occasionaly fail

Version:

v2.3.3

Date:

2024-09-19

Added

• Defaulted to allowing 50 sensors and increasing the number only requires docker-compose changes and making sure the firewall allows the ports. Lidar and IMU port ranges are now 9000-9049 and 9500-9549 respectively (previously 9000-9007 and 9500-9507).

Fixed

• Fixed a bug where large TCP messages were clipped

Version:

v2.3.2

Date:

2024-09-08

Fixed

• Fixed a bug where large TCP messages were clipped

Version:

v2.3.1

Date:

2024-08-08

Fixed

• Fixed settings migration issue when migrating settings from custom profiles

Version:

v2.3

Date:

2024-07-15

Added (Lidar Perception)

• New security profile optimized for crawling and running people detection

• Ability to manually start Event Recording from Viewer

• Alert in Viewer when no backend services are reachable

• classification_ts, sub-classification_ts and sub_clasification_ts TCP json stream

• JPEG web streaming when on full frame-rate

• Support for configuring host DNS settings through the Agent REST API

Updated (Lidar Perception)

• Disabled Zone Occupation Type and Point Count visibilities by default

• Disabled publishing of cluster points by default to save bandwidth. Cluster points will need to be enabled manually for Distributed Perception use cases

• Reduced extra decimal places in json output

• Allow users to override 20m constraint when setting the ML AOD

Fixed (Lidar Perception)

• System won’t try to reconnect to a sensor that has a UDP destination different from the current system. Instead, an Alert issued

• Corner cases where heuristic classifier wouldn’t change large objects back to unknown

• Fixed velocity output if frames are dropped

• Issue during offline .tar installations where local .deb packages were ignored

Added (LidarHub)

• New Perception fields: largest_arrival_delta: float, lowest_frame_rate: float, mean_frame_rate: float

• New lidar fields: percentage_incomplete_frames: float, tilt_since_reset: float

• Ability to trigger zone-based User Recordings (‘zone_id’ or ‘zone_name’)

• Ability to add an offender to a User Recording (‘offender_id’ == object.id or ‘offender_id’ == object.uuid)

• Optional ‘player_enrichment’ field added to User Recordings

• New object_list fields: num_failed_returns: int, classification_ts: int, sub_classification: str, sub_classification_confidence: float, sub_classification_ts: int

Updated (LidarHub)

• Default to no TLS and removed configuration options where applicable

• Removed use of wait_for_it (waiting for sensor discovery)

Version:

v2.2

Date:

2024-05-03

Added (Lidar Perception)

• Performance improvements (increased range)

• EULA in the product UI

• Added support for OTA updates

• Add REST endpoints to update agent from ouster.registry.com

Fixed (Lidar Perception)

• get-detect.sh script now confirms proper docker and docker compose version as well as presence of NVIDIA container run-time if Deep-Learning profile is set

• Fix post installation steps for RHEL package

Version:

v2.1

Date:

2024-02-29

Added (Lidar Perception)

• Added a zone height filter setting, to help reduce false positives

• Detect will re-initialize lidars found in an error state for a robust detection performance

• Added ability to visualize and edit each ML AOI (Area of Interest) through Detect Viewer

• The initial ML AOI for sensors defaults to sensor origin

Updated (Lidar Perception)

• Added a new Alert to detect incomplete lidar data (frames) indicating a network congestion issue

• Alert settings can now be changed without restarting Detect

• Optimized multi-sensor scenes for Deep Learning inferencing

Added (LidarHub)

• Added an ability to control the resolution of the background point cloud being displayed on the event recorder files

• Diagnostic stream now has host name, for easy management/identification

• Added Event Recording triggers based on zone point counts

• Automatic metadata is added to the event recorder files for better context

• Added support for having multiple Event Recording Configurations per zone

• Added more configuration options to keep the event recording length to manageable levels, especially on long events

• Zone point count information is reported in the Zone Aggregation data stream

Updated (LidarHub)

• Settings iteration bumped to versioning revision number (i.e. 2.1.0.X)

Fixed (LidarHub)

• Added timers to avoid any race conditions when various processes are getting rebooted ahead of the Ouster Solutions Agent

• A success response code of 200 is returned when manual event recording is stopped

• Added retries to the purging of MQTT publisher pending messages exceeding a set threshold

• mqtt_echo connection is left open on “Disconnect” to address any intermittent network issues

• Resolved a bug where the Event Recorder skips collecting JSON data when there is no flatbuf data

• Crash in Event Recorder Vacuum when compressed file incomplete

• Added corrupted settings file detection and graceful handling

Version:

v2.0

Date:

2024-01-03

Added (Lidar Perception)

• Added deep-learning inference models for detection and classification on supported NVIDIA GPU hardware

• Added support for parking lot occupancy detection application which doesn’t rely on background fade time

• Added support for Redhat, Fedora, and Amazon Linux 2 operating systems

• Added support for ARM-V8 based Compute hardware in addition to the already supported Intel/AMD compute chipset

• Added role-based access control (RBAC) allowing fine-grained control of user access by customer admins

• Introduced support for activating software licenses in offline deployments

• Exposed lidar angle, voltage, current consumption and temperature in source list on the Detect Viewer (Diagnostics tab). Angle is also on the Setup page

• Support for new settings profiles: its_settings, people_settings, dl_its_settings, dl_people_settings, and dl_parking_settings. Default profile is now people_settings. Profiles with dl_ prefix utilize deep-learning inference for detection

• Ability to read sub-classification in Detect Viewer

• Added ability to retrieve logs from the Detect Viewer when on 3rd party hardware

Removed (Lidar Perception)

• Removed endpoint for changing the password in Detect with introduction of RBAC

Updated (Lidar Perception)

• Disable refind module in people_settings and its_settings

• Changed default lidar and IMU port ranges to 9000-9007 and 9500-9507, respectively. When updating from earlier versions, existing lidar and IMU port ranges will not be updated

• Support for “include_classifications” field in event zone metadata to specify the classes that should trigger occupations

• Ability to toggle points count and occupation labels for event zones in Detect Viewer

• Changed default fade time in background filter from 3.3 to 10 minutes

• Changed empty object and occupations lists from TCP server to empty arrays

• Changed layout of the Settings page to be more intuitive. Moved the “Restart” button to be under “Services” panel on the Diagnostics page

• Exposed agent logs through Detect Viewer

• Ability to configure network interfaces through the Detect Viewer

• Ability to view firewall rules and perform ping tests from edge device through the Detect Viewer

• Ability to update all Gemini Edge software through the Detect Viewer

Fixed (Lidar Perception)

• Fixed issue where licensing alerts would remain active after successfully activating a license

• Fixed a bug that resets the underlay position when the opacity is changed

Added (LidarHub)

• Added Extrinsics to Perception Server Diagnostics for transmission over TCP sockets or MQTT protocol

• Support of Zone.has_point_count field in the zone definition

• Added Swagger tag descriptions for lidarhub APIs

• Lidarhub will collect and report Compute architecture (arm vs amd)

• Lidarhub will collect and report GPU stats via GPUtil/tegrastats for transmission over the diagnostics data stream

• Added support for Gemini ITS Edge module

• Added “world” long/lat settings to the diagnostics stream

• Added RPC on MQTT diagnostics stream (/[topic]/rpc)

• Added an alert when Event Recordings are not uploaded due to minimum available disk space exceeded

Updated (LidarHub)

• Made the suppression of empty object_lists and occupations optional

• Revised how outbound data rates are collected, calculated and reported in diagnostic data stream

• Change “world” API resource to use a payload vs query parameters

• Updated flatbuf schema to Detect 2.0.0

• Event Recording upload to Gemini Portal is made optional, after it is written to the local disk.

• Event Recording purge from the local disk after upload to Gemini Portal, is made optional

• Collecting and reporting CPU temp for Tegra-based devices via tegrastats, on the diagnostics data stream

• Allowed spaces in the list of TCP/MQTT sources

Version:

v1.3

Date:

2023-08-11

Added (Lidar Perception)

• Added a new diagnostic to detect sensor blockages

• Added 2D underlay such as a building layout or google maps screenshot in the Detect viewer tab, for better situational understanding of detections

• Added programmatic setup for single sensor install

• Feature to reset the sensor pose during initial install or to account for physical movement or the sensor

• Added PUT endpoint for sensor IMU pose

• Added Point Count Event Zones to understand zone occupancy for unknown object types for parking lot occupancy detection use cases.

• Added newer algorithms to improve multi-sensor alignment quality

Updated (Lidar Perception)

• Improved tracking performance to address objects getting temporarily occluded by obstructions in the scene

• Improved performance when the sensor experiences windy conditions

• Improved detection sensitivity

Fixed (Lidar Perception)

• Fixed the bug that crashed in background filter when changing settings that require a reset

Removed (Lidar Perception)

• Removed the websocket on port 3006 (“/node_stream”). We now publish raw node (align) clouds on port 3005, along with the sensor align node clouds.