99% LiDAR Detection Accuracy in Columbia County,  FL.

In Columbia County, FL, Seyond performed and accuracy test comparing LiDAR detection against inductive loop detection. ​

  • Context and Location

  • Technology

  • Methodology

Close cars

Comparing LiDAR and AI with inductive loops

LiDAR sensor technology offers a revolutionary, non-intrusive method for detecting all road users in real-time for better actuation and enhanced safety. ​
For cities, planners, and traffic operations teams considering a transition to a more cost-effective and efficient alternative, the accuracy and reliability of the data and analytics provided by LiDAR must be thoroughly verified.​

This is why Seyond set out to confirm the accuracy of its SIMPL solution using two sensors and a sophisticated perception software. This test was done to compare the accuracy of Seyond’s SIMPL solution against two existing inductive loops.

Testing took place in Lake City in Columbia County, Florida. The solution is installed at a standard intersection with three lanes in both directions. 
Lidar close up_7x11.5

LiDAR sensors and edge box

Two Falcon Prime LiDAR sensors running on FW4464 were installed on a cement pole at the corner of the intersection along with a POC edge box running our perception software (v1.3). This set up was specifically tailored to address the unique conditions of the location. Due to the difficulties in accessing suitable wiring to power the LiDAR sensor at another cement pole, both sensors were mounted on the same pole. This arrangement offered a comprehensive view of the intersection, although it also increased the risk of occlusion.

Inductive loop_7x11.5

Comparing two technologies against manual visual count

To assess the accuracy of SIMPL, a camera installed above the lidar sensors recorded road traffic to act as a ground truth. The footage was used to review discrepancies between the inductive loop and SIMPL​

To ensure a fair comparison, Seyond's team placed the virtual loops in the exact same location as the inductive loops. 

This testing was done from 2024-04-22 to 2024-04-23, involving 2 hours 50 minutes of recording analyzing two inductive loops, which is the equivalent of 5 hours 40 minutes of loop analysis.

RESULT SUMMARY

The LiDAR-based solution had a 99% accuracy rate

In total, 909 events were detected during the test through a visual manual count, acting as ground truth. The total count accuracy of the LiDAR solution was 99% and Seyond's solution proved to be more accurate than the inductive loop.

Result summary overview

 

SIMPL had fewer discrepancies than the inductive loops when comparing to the ground truth

Result summary discrepancies-1

Reliable system

From direct integrations with card networks and banks to checkout flows in the browser, we operate on and optimize at every level of the financial stack.

50+ integrations

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et.

Intelligent optimizations

Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.

Intuitive design

Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

RESULT EXPLANATION

Inductive Loop Discrepancies

1. Close Cars

135 misses were due to cars being close together in the inductive loop area. In these cases, the inductive loop counted them as one vehicle and could not discern between individual vehicle counts. The image above is an example of such a miss. The three vehicles that turn left actually only register as one long event from the Inductive loop.

2. Log ID Inconsistency - Attached-Trailer

The inductive loop counted a vehicle with an attached-trailer as three box ID. Seyond's solution counted only one.

3. Wide Turn Count

The inductive loops caught 38 events where a vehicle coming from the west side did a 'wide turn' to go north. As the car briefly went in the inbound lane and over the inductive loop, it was counted as a approaching car. The Virtual Loop solution did not log these 38 wide turn events.

 

LiDAR Solution Discrepancies

1. Lane Change Misses

Among vehicular misses by SIMPL, there were two lane change events which were not picked up by the solution but were picked up by the inductive loop. This could be remedied by making the loop size bigger or moving the loop to a better location.

2. Occlusion

There were two occurrences where there was Log ID non-tracking. One of them was due to occlusion when a larger vehicle came in between the LiDAR sensor and a smaller vehicle and the solution counted the smaller vehicle twice​, once before it was occluded and once after. This issue has been resolved thanks to better prediction and tracking algorithms to reduce the effects of occlusion.

In Summary

Overall, the accuracy of the Lidar solution compared to the inductive loop is nearly perfect but still requires some adjustment. The Seyond Team also plans to continue with a series of tests to investigate further the accuracy of the Virtual Loop solution in adverse weather and lighting conditions. 

Seyond is a leading global provider of image-grade LiDAR technology, powering a safer, smarter and more mobile world across the automotive, intelligent transportation, robotics and industrial automation sectors. Founded in Silicon Valley with strategically placed research and manufacturing facilities across the globe, Seyond is crafting LiDAR solutions that elevate autonomous driving and fuel the advancement of smart infrastructure development.

​