Skip to content

Appendix 3. Description of the parameters of the configuration file «TrackEngine.conf»#

Parameter Description Possible values Default values Editing recommendation
Logging modes CARS Stream (logging)
severity List of logged information that needs to be logged. - 0 – log warnings and errors, 2
- 1 – log system warnings,
- 2 – log only errors.
mode Logging mode. - l2c – show information only to the console, l2f
- l2f – show information only to a file,
- l2b - show information to both file and console.
General Parameters (other)
detector-step The frequency of detection of vehicle and LP in a given area before performing redetection. 0...30 7
detector-comparer Comparing vehicle and LP detectors to choose the bestshot. - 0 - DctConfidence, 1
- 1 - DctCenter,
- 2 - DctCenterAndConfidence,
- 3 - DctSize
use-one-detection-mode Choice of detection mode - 0 – unlimited number of detections per 1 frame; 0
- 1 – one detection per frame.
skip-frames The number of frames that should be skipped if there were no vehicle detections on the frame. If the detector did not find anything, frames will be skipped to save resources, after the specified number of frames, CARS Stream will detect the vehicle again. 0...50 36
frg-subtractor The mode of accounting for movements in the frame. Subsequent detection of the vehicle and LP will be performed only in areas with movement. Enabling this option increases the performance of CARS Stream. - 0 – on; 0
- 1 – off.
frame-buffer-size The number of frames in the CARS Stream framebuffer. Doesn't affect performance. If the buffer size is reduced, frames may be dropped on the RTSP stream. 10...1000 10
amount-of-free-space-in-buffer Amount of free space in CARS Stream frames buffer (measured frames count). 1...1000 32
wait-other-streams-for-processing-time Amount of time for waiting, while other streams become ready for strategy processing (measured in ms.). 10...1000 500
callback-buffer-size Buffer size for accumulation of information transfer functions to external systems. 10...1000 300
max-processing-fragments-count The maximum number of fragments for the stream processed in a specific time. (0 – no restrictions; more than 10 – take all frames from the stream) 0...10 600
batched-processing Shots batch processing - 0 – batch processing off 1
- 1 – batch processing on
min-frames-batch-size The value of the minimum frame batch size for processing. Higher values result in greater processing latency, but increase throughput and device usage. Zero/negative values disable this feature, so any streaming frames will be processed if they are available. The parameter is not used in current functionality. - 0
max-frames-batch-gather-timeout The maximum available waiting time to collect the next batch of stream frames from the start of the last processing (measured in ms). The parameter is not used in current functionality. - 0
parallel-tracks-processing To process tracks created after detection and redetection in parallel or all together. - 0 – in parallel 0
- 1 – together
frg-regions-alignment Fragments alignment. This setting increases the performance. 0 or negative values mean the usage of non-aligned areas. - 0 – off 0
- 1 – on
frg-regions-square-alignment Making fragments all the same size. - 0 – off 0
- 1 – on
detector-scaling Frame scaling before processing. Needed to reduce processing time. - 0 – on 0
- 1 – off
scale-result-size The maximum frame size after scaling to the largest side of the frame. If the original frame had a size of 1920x1080 and the «scale-result-size» value is 640, then the frame resolution will be changed to 640x360. The detection time decreases as the frame resolution decreases. 320...1280 640
max-detection-count The maximum number of detections per frame. The larger this parameter, the more CARS Stream vehicles will be able to detect in parallel. The increase is necessary for busy roads with heavy traffic. 1...1000 20
minimal-track-length The minimum number of frames in one track. 1...1000 1
tracker-type Tracker type - kcf; carkalmanV2
- opencv;
- carkalman;
- carkalmanV2
- vlTracker;
- none
kill-intersected-detections The parameter is responsible for the destruction of overlapping detections of different vehicles. Allows you to avoid jumping from one vehicle to another within the same track. - 0 – do not remove intersected detection; 1
- 1 – remove intersected detection;
kill-intersection-value The parameter is responsible for destroying one of the tracks when the IOU is greater than the set value. 0…1 0.6
build-intersection-value The parameter is responsible for matching the BBox with the track when the IOU is greater than the set value 0…1 0.3
Vehicles parameters (vehicle)
best-shots-number-for-track The number of best shots in one track. 1...1000 5
Faces tracking parameters (Face)
face-landmarks-detection Flag to enable face landmarks detection. Disabling it improves performance - 0 – detect landmarks; 1
- 1 – do not detect landmarks;
Pedestrian tracking parameters (Human)
human-landmarks-detection Flag to enable human landmarks detection. Disabling it improves performance - 0 – detect landmarks; 1
- 1 – do not detect landmarks.
remove-overlapped-strategy Parameter determining the strategy, used for removing overlapped detections after (re)detect. None – do not remove anything; both – remove both; score – remove the one with the little score. - none; score
- both;
- score.
remove-horizontal-ratio Width to height ratio threshold, used for removing horizontal detections. Horizontal are the detections where width is greater than the height. If the value is higher than the set value, the system will determine the detection as horizontal and perform its removal. This parameter is used only as part of the detection of people. 1.000…2.000 1.6
iou-connection-threshold IOU value threshold, used for matching tracks and detections 0…1 0.5
reid-matching-threshold reID value threshold, used for matching tracks to each other when breaking. Only used for humans. 0...1 0.85
non-active-tracks-lifetime Lifetime of old non-active tracks, that used for reID. It's measured in frames count and used for matching tracks to each other. Higher values lead to better quality of reidentification, but reduce performance 0...1000 100
reid-matching-detections-number Number of detections required to re-identify the track 0…100 7
use-reid Flag to enable usage of re-identification. - 0 – off; 0
- 1 – on.
frames-to-kill-track Number of frames with no pedestrian detected to complete the track.
This parameter can only be used when use-reid parameter is off. 1…200 4
Detectors parameters (detectors)
use-face-detector Enabling the face detector. Experimental mode. - 0 – off; 0
- 1 – on.
use-body-detector Enabling the body detector. - 0 – off; 1
- 1 – on.
use-vehicle-detector Switching on the operation of the vehicle detector. - 0 – off; 1
- 1 – on.
use-license-plate-detector Switching on the operation of the LP detector. Turning it off will cause system operation error. - 0 – off; 1
- 1 – on.
Frame proccessing strategy
strategy A strategy for maintaining tracks and sending the best frames to CARS Stream using the TrackEngine library. - common; coroutineV2
- redetect;
- droi;
- droiFgs;
- lpDroi;
- coroutine;
- coroutineV2
strategyPriority/Rank Best detections selection priority. - common – biggest sized detection is considered the best bottom
- bottom – detection located the closest to the bottom of the frame is considered the best
BottomRankDelta The minimum number of frames to the most recent for prioritization to start working 10...100 40
redetect-merge-intersection-detection-threshold Threshold value for merging similar detections 0…1 0.5
Debug
save-debug-info CARS Stream debug mode. - 0 – on; 0
- 1 – off.
show-profiling-data Output log of CARS Stream operation. - 0 – on; 0
- 1 – off.
save-buffer-log Saving buffer operation logs. - 0 – on; 0
- 1 – off.